Wednesday Nov 07, 2007

Processors and Performance : Chips, MIPS, and Sizing blips..

The following post is a close proximity of an article that I published in this month's Sun "Technocrat" (November 2007) issue. Hopefully you'll enjoy this discussing regarding the past and present relationship of CPU's and architecture to system performance.

In today's fast paced world of ever increasing demands for system throughput, the foundation of discussion and expectations typically all hinge upon the same topic.. CPU performance.   This article will be an examination of CPU's and system architecture, as they relate to performance and capacity planning as a whole.   From our last discussion "The Many Flavors of System Latency ..." , we will extend the context to focus on past and present competing aspects of system/CPU architecture,  including a brief history of how we got to the current competitive landscape we find ourselves in today.  (The photo to the left is of a Sun T1 "coolthreads" 8 core, 32 thread cpu)


Going back to the early days of microprocessor design (and also likely a familiar topic from your Computer Science Bachelor's curriculum), you will recall much conversation, speculation, and competition surrounding two competing approaches to CPU architecture :  CISC vs. RISC.

In a nutshell, CISC (Complex Instruction Set Computers) designs emphasize the use of "complex" instructions within the HW to minimize the amount of Assembly code (SW) required.  Other benefits are that compilers don't need to be as complex, as well as CISC CPU's requiring less RAM to store instructions.  However, this approach sometimes requires more than one clock cycle of a processor to complete processing a complex instruction.

RISC (Reduced Instruction Set Computers), just as the name refers, offer a reduced number of simple instructions that can complete execution within a single clock cycle, but which might require multiple instructions for a complex operation such as "multiply".  The RISC approach has the nice side-effect of also requiring less chip "footprint" (in # of transistors, etc..)  reserving more die space for memory registers, while also typically offering streamlined execution within the same window of execution time as CISC counterparts.  Modern compilers (such as Sun's Studio 12 line) offer significant performance benefits that should not be overlooked, especially critical when developing and compiling binaries on RISC based architecture (Sun has seen performance optimized benefits of 200+ % when using the latest releases of Sun Studio Compilers vs. generic "gcc" compiled code).

Today, the market is segmented in more or less the same 2 camps, but they are divided down the lines of "x86" compatible CPU's (modern CISC designs from the Intel and AMD) and the modern RISC competition from Sun (SPARC) and IBM (Power), where HP and DEC (now gone altogether) have stepped back from RISC manufacturing.

Is there more to Moore's Law ?

For nearly the past 30+ years, we have seen processor performance double according to Moore's Law (approximately every 2 years), a movement from Uniprocessor based systems to those that required scalability vertically, into multi-processor systems that we've become familiar with in the Unix world, better known as SMP (Symmetric MultiProcessor) systems.   It's funny that Gordon Moore made his original claim in a 1965 article of electronics magazine originally regarding the trend of Integrated Circuit component counts doubling every year, only loosely predicting this trend might continue until 1975, though he was uncertain that any future projections could be made (he later changed his prediction in 1970 to "doubling every 2 years", which has held relatively steady ever since).   Since past trends of IC / CPU transistor counts have closely correlated to CPU performance, the association of Moore's Law to performance  was made.

In order to accommodate this rapid rate of growth in processor performance, the number of transistor's contained within a CPU has been just one of the key characteristics that have climbed (in addition to clock speeds, etc..).   Amazing as it may sound, but 45nm (nano-meter) manufacturing fabrication is expected by Intel and others hopefully in 2008 (when just 10 years ago we had 500nm manufacturing) .  Even at the levels that we were at in 2004 with 90nm manufacturing, the width of a transistor (50nm across) was 1/2 the diameter of a single celled influenza virus !   Given that we are approaching atomic dimensions of gate thickness (at < 1nm), the pace of transistor density and clock frequency increases (using current designs) appears to be approaching the physical limitations of manufacturing.  In addition to these concerns, power and related heat issues have become very pronounced in today's "green" computing campaigns.  Luckily, Sun has dealt a hand worthy of industry recognition that includes CPU innovations to keep us progressing forward, however addressing more than simple "transistor counts", but more aptly the efficiency of moving an SMP  like architecture "onto" a piece of silicon, hence.. a system on a chip (which is essentially what we have with the T2).

How many ways can you weave those THREADS ?

SMP systems required an Operating System kernel that had a means of fairly sharing the CPU resources among the processes awaiting execution within the system (via priority-based scheduling classes).  This capability within Unix for "time slicing" between "runnable" processes and available kernel and physical CPU resources on systems hinges on the kernel dispatcher associating processes with "light weight processes" that in turn get bound to kernel "Threads", in order to be run within the "context" of physical processor registers and an execution pipeline (aka, HW Threads).  For the most part, this is how today's Solairs OS functions, along with the appropriate dose of preemption and locking mechanisms.

Over the past decade, and along with the increased demands of internet traffic (and associated application workloads), applications have gradually become better able to scale vertically within systems, primarily through the use of SW multi-threading.   Within a multi-threaded application, many threads of execution can run simultaneously across available CPU's within a system, allowing for an application to scale as close to "linearly" as possible (doubling the application throughput as the # of available CPU's doubled).

    THREAD (of Execution) :
  • noun          One of many software "threads" of execution that can be processed simultaneously on a computer system.
    Linear Scalability :
  • noun          The ability to increase system performance (throughput) at the same rate that resources (cpu's,..)are added.

CPU's and Memory :  UMA and NUMA

Modern computing systems also offer a "shared memory" model that has very specific performance and latency related characteristics, depending upon the system design (physical system interconnect type [bus/crossbar], memory management controllers, proximity to physical cache/RAM, etc..), the Operating System Kernel memory management (memory mgmt libraries, JVM Garbage collection, etc..), as well as the Application SW execution characteristics and memory requirements (cache hit ratio's [Instruction vs. Data Cache], TLB miss characteristics, RAM requirements, ..).

Among the modern types of vertically scalable parallel computer architectures available for multiprocessor systems, one of the most common high performance designs has become NUMA/ccNUMA (Non Uniform Memory Access / cache coherent NUMA).  Sun's E25K systems fall within this category, as do most large systems that offer vertical scalability of many independent CPU / Memory boards, acting together and communicating across a shared system interconnect (backplane/centerplane/crossbar).

Previously, local proximity of memory and the latency associated with accessing that memory was uniform and predictable.  With the advent of system growth beyond single system (cpu/memory) boards, UMA (Uniform Memory Access) could no longer be guaranteed.  One common performance issue that must be addressed within large NUMA environments is the aggregate impact of both physical memory proximity, alongside the gap between processor speed and memory latency (see below).   Solaris addressed this issue (with a Solaris 9 update) by introducing MPO (Memory Placement Optimization) that associates memory physically closer to a cpu to minimize the additional cross interconnect memory latency (this couples Solaris and the underlying HW along with Cache Coherency within ccNUMA architectures).  Note, other optimizations have been used to address large Sun Enterprise system centerplane latencies, including kernel cage splitting, removal (if DR isn't required), and the introduction of S10 enhancements.   <\*"busstat" can be used to diagnose these issues\*>

Lucky for us, we are quicly approaching system architectures that once again allow for UMA designs, offering memory access predictability (with the T2 and related offerings moving forward, where density of cpu cores/threads brings greater multithreaded capacity within a small footprint).  However, the current state of the industry isn't quite as lucky if you look across the bow of the competition, and within our client production environments.

The growing CPU - Memory gap ...

As you can see from the diagram to the left, over the past several years as microprocessor design has moved rapidly, doubling the performance and clockspeed of CPU's roughly every 2 years, the same increases have NOT been matched within the Memory arena.  This "wait" time that threads of execution must incur, coupled with the additional latency required for electricity to travel greater distances to access memory across the system interconnect (not physically local to a CPU, but rather on another CPU board or memory bank) can impact processor efficiency and overall system / application throughput dramatically.   (this slide and the next are from )

Chip Multi-Processing + Hardware Multi-Threading = Chip Multi-Threading

Realizing the implications of the memory latency "lag" (and somewhat against the tide of relying upon ever-increasing CPU clock speed increases), several years ago Sun made the decision to address this with an acquisition of Afara Websystems to bolster it's processor lineup and chart the industry's new course toward multi-core CPU's.

From the diagram on the left, it can be seen that Sun's adoption of both CMP (US-IV/+), aka Chip level Multi-Processing (having multiple cores per cpu, each with an execution pipeline), alongside HMT .. Hardware Multi-Threading (adding a multi-threaded execution pipeline within a core/cpu), shores up and nearly eliminates the issue of idle CPU cycles waiting on memory operations (by offering many physical threads of simultaneous execution within a CPU).  This is the foundation of what Sun calls CMT (Chip Multi-Threading), which is reflected in Sun's CPU roadmap for both the Niagara (T1 and T2 CPU's), Rock CPU, as well as the Sun/Fujitsu Olympus CPU (as a follow-on to the US-IV+).

Why so much $cache ?

In order to further minimize system latency and ensure peak performance, modern architectures include Memory Management and CPU memory access mechanisms "on-chip", such as the L1 / L2 cache and MMU located in Sun's CMT products.  Below is a block diagram of Sun's latest "T2" SPARC Core Architecture (based on the 64 bit SPARC V9 instruction set).

    Critical components for optimal kernel CPU / memory performance:
          • Level 1 (L1) Data Cache / Instruction Cache ..
          • Level 2 (L2) Secondary Cache, on-chip and shared for Sun CMT processors
          • Level 3 (L3) <only Sun's US-IV+ cpu offers L3 cache on-chip>
          • I-TLB (L1 / L2) Instruction -Translation Lookaside Buffers
          • D-TLB (L1 / L2) Data -Translation Lookaside Buffers
          • Buffer cache (Filesystem Kernel "page cache", taken from the "free list" of available RAM; this is a kernel structure)
Note:  The kernel memory page "size" is determined in part by the CPU chosen, since x86/x64 CPU's
                (Intel/AMD) offer 4K pagesizes, while the UltraSparc I through IV... offer 8K / 64K /
                512K / 4MB page sizes.   (\*monitor with the pagesize, trapstat, cpustat, and pmap ..\*)

Examining the key attributes of your system's Workload as Requirements  :

Once again, in order to select the appropriate architecture for a production deployment, the all-inclusive entity that we need to examine in it's entirety is the "Application Environment", all of it's subsystems, as well as individual /concurrent workload characteristics :
  • Single Threaded vs. Multi-Threaded Applications (OLTP, DSS, HPC, Web/AppSvr, .. ?)
  • Compute / HPC Intensity (MPI, horizontally scaled compute farm requirements, ..)
  • Network I/O (long-lived connections vs. short-lived; large vs. small packets, # inbound RX pkts/sec..)
  • Memory Intensive Workload (shared / distributed memory req's, etc)
  • Storage Workload (R/W %'ages, Cache cfg, # Controller Interrupts/sec, # Files opened, shared FS,etc.)
  • Integer vs. Floating Point Calculations (T1's are not well suited to FP workloads)
  • 32 vs. 64 bit needs/benefits (address space needs beyond 4GB RAM ?)
  • Do SLA/SLC reqmt's focus on Throughput, BandWidth (IO/Net/Mem), Availability, and/or Response Times ?

Choosing the right CPU for your Workload :

Sun SPARC based CMP / CMT CPU's :

  • US-IV+ :  Well suited for very large vertically scaled configurations where the 32MB of L3 cache makes a big difference, such as DB Servers (many of these environments have large single threaded processes, including batch/OLTP).  Sun's E25K systems scale up to 72 CPU's per single domain (144 cores).
  • Sun / Fujitsu SPARC64 (US-VI Olympus) :  Follow-on to the US-IV+ CPU.  (~1.5\* performance of a US-IV+)  This CPU should have higher clockspeeds starting at 2.15GHz, with an additional HW thread/core, but no L3 cache on-chip.  Note that the up-coming CMT ROCK chip from Sun in 2008 will fall within this segment.
  • T1 (Niagara 1) :  Well suited for small to medium sized Multi-Threaded Workloads (that don't have much /any FP processing.  Best for : WebSvrs, AppSvrs, DNS, etc..).  Each single socket system offers up to 32 HW Threads of execution.
  • T2 (Niagara 2) :  Well suited for medium to large Multi-Threaded Workloads.  These systems should also offer good general purpose computing performance, given the addition of FGU's per CPU core, along with built-in 10G Ethernet, etc..  Each 5x20 system presently offers a single socket with up to 64 HW Threads.  Look for multi-socket systems based up on the T2 cpu in the not so distant future ;)

Sun's "World Class" T2 (Niagara 2) CPU :

At a glance, the new T2 CPU offering from Sun is a true "system on a chip" that lives up to it's reputation as "the worlds fastest processor" (the new world record benchmarks listed further down in this article can attest to the validity of that statement and show how Sun's latest CMT CPU's are changing the landscape of computing efficiency, part of the reason why Sun calls this "CoolThreads" and/or "Throughput" computing).

T2 CPU Highlights :
    • 8 cores \* 8 Threads each = 64 Threads of Execution
    • 65nm, initially running @ 1.4 GHz
    • 8 Floating Point / Graphics Units (FGU's, one per core)
    • on-chip Crossbar providing : 180 GB/s R + 90GB/s W
    • built in 2\* 10Gb Ethernet, MMU, Encryption, etc...

The following table provides a high level comparison of Sun's T2 and T1 CPU's  :

(for the complete Microprocessor Review report on the T2, click here)

(For further details comparing the T2 to other recent Sun CPU's for #transistors, etc., click here)
(For photos inside the new Sun T5x20 systems, based upon the T2 cpu, click here)

Solaris kernel (CPU related) Performance Metrics and Utilities :

The following table is only listed as a "high level" sample of common metrics frequently used as part of Solaris CPU-related performance analysis.  This is by no means a comprehensive list of metrics available, but rather an introduction for those that aren't familiar with the essentials.   An up-coming set of blogs will include much more detailed examples with command line output, also including discussions of kstat and Dtrace visibility available.

Note: \*  vmstat, cpustat, trapstat, intrstat  reflect system-wide statistics, while  mpstat, cputrack reflect per CPU statistics. \*

Run Queue
 Kernel Threads Runnable, but not executing   (best if 0, or at most < # cores)
vmstat (r)
Blocked Kthr
 Blocked Kernel Threads             (typically ID's an IO bottleneck, see also %wt, lockstat,.. )
vmstat (b)
System Calls
 Number of System Calls (calls made into the OS kernel, accounting towards %Sys)
vmstat (sys)
 Number of System Interrupts per interval (interrupts have the highest priority on the system)
vmstat (in)
% CPU (U/S/I)
 % CPU utilization (% User space / % System kernel / % Idle);  % User  should be  2\*  % Sys
Cross Calls
 Per CPU Cross-Calls (either for cross processor interrupts, and/or maintaining cpu virtual memory translation consistency .. aka cache consistency with MME and mapping TLB entries, etc.)
mpstat (xcal)
Cpu Interrupts
 Per CPU Interrupts   (also use intrstat, as well as lockstat for system correlation)
mpstat (intr)
Context Switches
 Involuntary context switching (icsw reflects preemption..) vs. voluntary context switching (csw)
mpstat (i/csw)
CPU Migrations
 Per Cpu Migrations .. A more inclusive migration off of and onto another CPU.
mpstat (migr)
Shared Mutex
 Mutex exclusion lock activity (per cpu)  p/lockstat gives the best visibility of this activity.
mpstat (smtx)
% CPU Waiting
 % of a single CPU spent Waiting (during the sampling interval).  See also  b kthr.
mpstat (%wt)
 Instr TLB Misses
 % of MMU related Instruction Translation Lookaside Buffer Misses (see also  pagesize, pmap, cpustat..)
trapstat -t/T
 Data TLB Misses
 % of MMU related Data Translation Lookaside Buffer Misses (see note above as pgsize is related)
trapstat -t/T
  CPU Counters
VARIOUS CPU specific HW event counters (Cache, Instruction level, FP, TLB; man cputrack for your HW specific counters available)
  CPU Counters
 VARIOUS System Wide CPU event Counters (man cpustat for your HW specific counters)
  BUS Statistics
 Available System Specific Bus Device / Instance Counters & Events (use busstat -l  for your HW)
 kernel Statistics
 ALL kernel statistics are available individually via kstat  (module:instance:name:class)

\*\*NOTE: if you'd like to try a single Solaris utility that can run in minutes to automate the performance / workload correlation and reporting for you, take a look at sys_diag if you haven't already done so already (or the README).  It includes both high-level (vmstat, mpstat, iostat, netstat, kstat, ...) snapshot and analysis, as well as Deep analysis mode which includes extended Dtrace /dexplorer and lockstat probing.   (all output is summarized and color coded in an HTML report header/ Dashboard with a Table of Contents for analysis details) \*\*

Common CPU Benchmarks and What they mean :

The following list provides a set of definitions and examples for some of today's most common independent (industry accepted) computing benchmarks.

  • SPEC CPU2006  :  CPU-intensive benchmark suite, stressing a system's processor, memory subsystem and compiler. SPEC designed CPU2006 to provide a comparative measure of compute-intensive performance across the widest practical range of hardware.  This benchmark suite includes both the SPEC int_rate2006 and SPEC fp_rate2006 benchmark tests.
NOTE that SPEC is an independent, non-profit 3rd party benchmarking organization, providing the comparative examples that follow.

IBM System p570
HP ProLiant DL360 G5

NOTE:  For the benchmark example above, and those that follow, this shows how Sun's latest CMT T2 cpu is changing the landscape of computing efficiency, part of the reason that Sun calls this "CoolThreads" and/or "Throughput" computing.
  • SPEC jbb2005 :  SPECjbb2005 (Java Business Benchmark) measures the performance of a Java implemented application tier (server-side Java). The benchmark is based on the order processing of a wholesale supplier application. The metrics given are number of SPECjbb2005 bops (Business Operations per Second) and SPECjbb2005 bops/JVM (bops per JVM instance). 

IBM p6 570
HP 2660
Dell 2950
Space (RU)
Power Consumption (Watts)
Performance (BOPS/JVM)
Performance / Watt
  • SPEC jAppServer2004  :  SPECjAppServer2004 is the only industry-standard benchmark used for Java Enterprise Edition application servers. In addition to testing application server performance, it also tests the database performance of servers deployed to support the application tier.
Database Tier, SPECjAppServer2004 2-Node Comparison Table

Dell 2900
Space (RU)
Power Consumption (Watts)
Performance (SPECjApp JOPS)
Performance / Watt

A word on SWaP

Given the state of environmental (global warming concerns), not to mention Power, Cooling, and floorspace costs, Sun has created the SWaP metric to compare and reflect the relative performance when taking into account the "space" (Rack Units), as well as "power consumption" (Watts).

The calculation is :      SWaP    =       Performance (operations or transactions per interval)
                                                            Space (RU)  x    Power Consumption (Watts)

Sun Benchmarks and Comparative Methodology :

For the purpose of internal only comparative benchmarking, Sun provides and maintains(internal-only) results for AMP v.2 and M-value benchmarks.

As stated very clearly at both sites (from the URL's noted above) :

The most appropriate way to size a Sun server for a specific application is to engage Sun's Competency Centers.  These centers will recommend a system using their knowledge about the application performance on Sun systems, and based on specific information from the customer.

Over the past 12 years with Sun, on several occations I've been brought into a mission critical production environment having performance issues just after going live.   The reasons which were most commonly the cause of this include :

  • NOT doing any type of actual Pre-Production Testing on the "target" configuration to be deployed, including :
    • NOT doing any sort of formal POC (Proof of Concept) with the target configuration to be purchased or migrated to.  A proof of concept is typically not an all-out formal benchmark effort, but would minimally give you the opportunity to run conduct Functional Testing, in addition to some simulated "production like" load tests against the configuration, that could demonstrate that the staged target environment will meet a  representative sample of "production like" workload.
    • NOT doing formal benchmarking, using a copy of Production data, and simulating actual samples of the most active production workloads (DB queries, Client Access patterns, Network traffic, etc..) with a tool such as LoadRunner.
  • While the reason above is nearly always the case, the cause that frequently crops up in most of these scenarios is that the only "Sizing" done was to compare the "before" and "after" M-Values to generate quotes !   I have encountered this even in mission critical environments, where NO pre-production staging or load testing was done ! This is just WRONG, it's more than a bad practice, possibly one that could get you fired if your (or your customer's) production environment goes down in flames after lots of $$ was spent !  Push back if necessary !
Use proper process and methodology in actually staging a new configuration and perform representative "production like" load testing.

\*\* Single-Core to Multi-Core (CMT) M-Value ("on paper") Comparisons Should CAUTIOUSLY be evaluated !! \*\*

Regarding the trend of migrating production environments from single-core to multi-core architectures, Beware!  Even though a generic benchmark test might reflect much higher #'s with fewer HW threads (and/or cores) than a current production configuration has, realize that there is a lot more to proper sizing and capacity planning (realizing that each configuration is unique) than is reflected in M/GHz (or in M-values) ! A lot can be said for certain types of workload requiring a specific # of HW cores for their environment to perform optimally "on cpu" without much cpu/kernel contention (locking, High TLB misses, context switching, and/or cpu-migrations..).

Hopefully this article has helped you reflect on the wide variety of CPU options available, as well as how they play such a significant role as the "cornerstone" of system architecture and overall performance of our customer's production environments.  Enjoy, and "let the chips fall (or rise) as they may"... :)

For more information regarding Performance Analysis, Capacity Planning, and  related Tools, see Todd's Blog at :

\* Copyright 2007 Todd A. Jobson \* 

Friday Aug 03, 2007

System Profiling 101 : Getting started using sys_diag v.7.04

The following entry is a variation of an article that I created for this month's Sun "Technocrat" publication (Aug. 2007).

This posting demonstrates the art of system profiling (from a high level overview) by introducing a few sample screenshots of the sys_diag  .html  report (it's header, Dashboard, and Table of Contents).. demonstrating how in a few minutes, sys_diag can present you with an accurately depicted system profile !

Note, the .html report snapshot samples presented here, match the command line output from my previous blog postings (from the same run of sys_diag).

If you haven't had the chance to try out sys_diag yet, this should give you the highlights of what you can expect in the .html report header sections.

Enjoy and let me know what you think,


Real world PROFILING ..

So.. what is Profiling.. ?? ... Well, in the real world, you can define profiling in many ways.

.. from the "profile" of the person standing next to you (what you see from your vantage point), to the personality "profiles" that we've all heard of in psychology (characterization based upon key attributes) ..

In your standard dictionary you'll find a definition such as this (from :


  •  the outline or contour of the human face, esp. the face viewed from one side.
  •  a verbal, arithmetical, or graphic summary or analysis of the history,
    status, etc., of a process, activity, relationship, or set of
    characteristics: a biochemical profile of a patient's blood; a profile of national consumer spending.
  •  a set of characteristics or qualities that identify a type or category of person or thing: a profile of a typical allergy sufferer.
  • Psychology. a description of behavioral and personality traits of a person compared with accepted norms or standards.

System Profiling .. in the World of Computing

Well, in the world of  technology, and more specifically.. Computing, "profiling" takes on it's own connotation, though similar to many of the more technical definitions noted above.

To some, system profiling simply includes a high level summary of resource utilization and bottleneck identification of a system during some period of data collection (point in time or over a duration).

"Broad Spectrum" System Profiling ..

System profiling to me is the characterization of a system as a whole, given a set of data, either for one event/point in time, or over a duration.  This characterization goes beyond workload (as you'll typically hear the term "workload characterization"), which is why I call it "Broad (or Full) Spectrum" profiling, more broadly taking into account and including :

  • Configuration characteristics (of all sub-systems/components within the "application environment" being profiled).
  • System Performance metrics captured, reflecting the variations in system/subsystem activity measurements (utilization, contention, throughput, latency, etc...).
  • Workload Characterization : Details correlating the Workload that was ongoing during the data collection (workload characteristics.. TPS/ Response Times/ Mbps/ ...), with the measurements taken.   (beyond the internal system workload identified, External sources need to be correlated)
  • A Characterized (Summarized) "Profile" of overall System Efficiency and "health" based upon Performance Analysis findings (system/sub-system Avg/Peaks.., Utilization vs. Workload, etc ...)
  • Notable Events and/or Exceptions encountered from the data available.
  • ... etc ...

"Narrow Spectrum" System Profiling ..

This would be in contrast to "Narrow Spectrum" (Focused) System Profiling, where attention to detail is focused in a very "narrow" and specific area of interest for analysis (typically in determining a Root Cause where a specific bottleneck is know within a sub-system or specific component of the system).

Note the common themes of defining Requirements, the "Application Environment", etc.. as presented in my previous postings .. (eg.  "What is Performance ? .. in the Real World", etc..).. and likely to be common themes.. in the Real World.. ;)

Look for more details on this and much more in an up-coming blog entry more thoroughly delineating the distinction between Profiling, Workload Characterization, Performance Analysis, and Capacity Planning....  

For now, enjoy the following discussion on how sys_diag can have you profiling in no time at all ... :)

Profiling with sys_diag ...

Automating Solaris Performance and Configuration Analysis

In the arena of Performance and Configuration Analysis, the freely available Solaris utility “sys_diag” offers the capability to automatically capture this information in a single .html report (also .txt and .ps) after running one easy to use ksh script. Typically, several utilities/tools need to be run separately, requiring manual collection, aggregation and correlation of the data, prior to conducting the analysis of data.

Sys_diag automates this legwork, by running over 100 Solaris built in commands/utilities (depending on the parameters used) and presenting the data as a structured report with a summarized header, a color-coded “dashboard” (broken down by high level workload characterization, sub-system findings, followed by a Table of Contents), all with links to the corresponding report sections with detailed configuration and/or performance analysis findings.

sys_diag 's  HTML Report Header :

sys_diag 's  HTML Performance "Dashboard" :

 The following sample .html system performance “Dashboard” (a portion of which is shown below) reflects the 4 key sub-systems (CPU/Kernel Profiling, Memory, IO, and Networking) as a summarized depiction of sub-system “health”, based against a list of rules / thresholds that the captured data is compared to during post-processing.

These rules and thresholds are listed in the Performance Analysis section (Section #24) and can be easily tuned / modified to offer more stringent or lenient identification of performance exceptions that contribute to the Green/ Yellow / Red (OK / Warning / Critical) color-coded status within each dashboard section. Within each section are listed the key performance metrics and a summary of exceptions, along with Average and Peak (High Water Mark) values present during the collection/sampling period. At the end of each section is a list of key “links” to the substantiating detailed data analysis within the report.  

\*Click to Enlarge\*

When run for performance data gathering (-g or -G), 2 types of performance data is captured :

\* vmstat, mpstat, iostat, netstat, .. data for a duration (-T total_secs), captured at specified sampling rates (-I interval_#secs). The default duration is 5 minutes of data capture @ 2 second intervals if -I / -T are not specified.

\* 3 Point in Time detailed snapshots (beginning, mid point, end point). If -G is used, and         Solaris 10 is the OS, then Dtrace and detailed lockstat, pmap/pfiles, cputrack, ...  snapshots will be taken (beyond the core “-g” snapshots that include  ps, netstat -i, vxstat, kstat,  ...).

Sys_diag has been run on virtually all models of Sun systems running Solaris 2.6 or > (from x86 laptops up to fully loaded E25K's), offering extensive Solaris 10 configuration and performance data, including DTrace snapshots. It creates a single .tar.Z compressed archive (including all raw, snapshot and post-processed datafiles) that can be emailed/ ftp'd.. for performing system configuration and/or performance analysis off-site.. from virtually anywhere.

This is one of the key characteristics that sys_diag offers.. to save a LOT of time.. not requiring many separate manual runs / collection / correlation of data, or the need for any 3rd party tools, libraries, or agents to be installed on a system other than downloading the "sys_diag" ksh script itself. Virtually no learning curve is required for loading, running, and reflecting basic performance profiling, including high level sub-system bottlenecks (deeper root cause correlation might require some level of advanced system administration knowledge, though virtually all the data needed will have been already captured by sys_diag).

This utility has been used extensively in the field over the last several years, run on literally hundreds of production systems as part of escalation root cause analysis, in addition to providing the basis for dozens of Architectural and/or Performance Assessments
(including formal Capacity Planning / Benchmarking). Graphing of the data captured (vmstat, netstat...) is also easy to do using StarOffice as explained in the README file that sys_diag creates.

sys_diag 's  HTML Report "Table of Contents" :

The screenshot below shows the Table of Contents and related sections available within the .html report (\* Click to Enlarge \*) :


Although this tool isn't meant to replace long-term historical Performance Trending and Capacity Planning packages (Teamquest, etc..), it provides the foundation and basis for a very robust starting point (and actually is much better at point in time workload characterization and root cause analysis of bottlenecks, where very granular detailed data correlation is required).

Over the time that sys_diag has been posted on BigAdmin, many Sun customers around the globe have downloaded and commented positively on their experiences with it. For more information, or to download and try it out for yourself , the following URL's should help you get started :




The latest release of sys_diag (v.7.04) is available from BigAdmin
(unpackaged ksh) at :

sys_diag  is also available as part of the "SunFreeware" Distribution
(packaged with the README) at :

The following recent blog postings provide an extended overview of sys_diag and it's capabilities :

Solaris Performance Analysis and Monitoring Tools... at what cost ?...

What is sys_diag ?? .. Automating Solaris Performance Profiling and Workload Characterization.
sys_diag v.7.04 command line output ...


\*\*Note, read the ksh script header pages or the README file prior to using, and ALWAYS test first on a representative non-production system.. as is the best practice when making ANY production environment changes... ;)

(Copyright 2007, Todd A. Jobson) 

Add to Technorati Favorites

Monday Jul 30, 2007

sys_diag 7.04 on !

As of 7/30/2007, will now be including the Solaris utility "sys_diag" as part of their distributions (Solaris 8 -> Solaris 10 for both Sparc and x86).

 The format at SunFreeware varies BigAdmin's distribution, only in that BigAdmin provides the raw ksh script, where sys_diag on SunFreeware is packaged along with it's README file in the standard Solaris package format (pkgadd). :     sys_diag v.7.04 : README_sys_diag.txt

If you're not yet familiar with, you should be ! .. check it out asap for a great selection of Solaris freeware (already compiled and packaged for you !!).

Tuesday Jul 17, 2007

sys_diag v.7.04 command line output ...

The following output was captured recently from running sys_diag v.7.04 on a (Solaris 10u3) Sun Ultra60 2 cpu test system in my lab  

Note the list of utilities run and types of data captured, as well as the final performance summary (a small summary of the complete color coded HTML dashboard available in the full .html report).

sys_diag has been run on virtually every type of Sun system, running Solaris 2.6 -> S10.  I have personally conducted dozens of Performance Analysis, Capacity Planning/ Benchmarking, and/or Architectural Assessments using sys_diag in production environments.. x86.. up to fully loaded E25K environments.

The latest release of sys_diag (v.7.04) is available from either BigAdmin (unpackaged ksh) or (pkg'd with the README) at the following URL's :


Realize that more than half of sys_diag 's benefit is in working from the .html aggregated report file.. that links and correlates all the independant data files together with findings and exceptions via a nice color-coded header / dashboard / and Table of Contents.  (the legwork is all done for you !)

I'll try to get a sample snapshot of a report header/dashboard for an up-coming blog... but for now, just download and test run sys_diag (v.7.04 is recommended), review the final .html report and forward and questions/comments back to me.. along with RFE for future releases.

(Read the last sections of the README for a detailed description of all datafiles created/available...)

With a little practice, it should save you many hours.. if not days.. of effort as it does for me.

Enjoy and let me know what you think,  


The following example does the deepest level of Performance data Gathering (-G, which includes Dtrace and pmap/pfiles snapshots vs. -g for light-weight perf gathering), Verbose output (-V), in addition to creation of a long/detailed configuration report (-l). The sampling rate used is 1 second intervals (-I1) for a total duration of 298 seconds (-T298).

\*Without -I || -T, the defaults are 2 second samples for 5 minutes total data gathering. Also note that when -G && -V are used together, the initial Dtrace and Lockstat snapshots take a couple minutes to complete, prior to beginning the data collection for 298 seconds (since the duration of probing is expanded with -V to 5 seconds vs 2 seconds with -G alone, or 1 second minimal lockstat sampling using  -g  ..aka.. no Dtrace probing).

root@/var/tmp #  ./sys_diag -G -V -l -I1 -T298

sys_diag:0717_033209: GATHER Extra PERFORMANCE DATA (-G)
sys_diag:0717_033209: VERBOSE (-V)
sys_diag:0717_033209: INTERVAL : 1 second sampling (-I1)
sys_diag:0717_033209: TIME Duration: 298 seconds (-T298)
sys_diag:0717_033209: LONG report (-l)
sys_diag:0717_033209: # Creating ... README_sys_diag.txt ...

sys_diag: ------- Beginning Process SNAPSHOT (# 0) -------
sys_diag:0717_033209: Dtrace: TCP write bytes by process ...(_dtcp_tx Snap 0)
sys_diag:0717_033209: Dtrace: TCP read bytes by process ... (_dtcp_rx Snap 0)
sys_diag:0717_033209: Dtrace: systemwide IO / IO wait... (_diow Snap 0)
sys_diag:0717_033235: Dtrace: Syscall count by process... (_dcalls_ Snap 0)
sys_diag:0717_033243: Dtrace: Syscall count by syscall... (_dsyscall_ Snap 0)
sys_diag:0717_033251: Dtrace: Read bytes by process... (_dR_ Snap 0)
sys_diag:0717_033258: Dtrace: Write bytes by process... (_dW_ Snap 0)
sys_diag:0717_033306: Dtrace: Sysinfo counts by process... (_dsinfo_ Snap 0)
sys_diag:0717_033314: Dtrace: Sdt_counts ... (_dsdtcnt_ Snap 0)
sys_diag:0717_033321: Dtrace: Interupt Times [sdt:::intr].. (_dintrtm_ Snap 0)
sys_diag:0717_033321: # ps -e -o ...(by %CPU) ... Snapshot # 0
sys_diag:0717_033321: # ps -e -o ...(by %MEM) ... Snapshot # 0
sys_diag:0717_033332: # pmap -xs 519 ...
sys_diag:0717_033332: # pmap -S 519 ...
sys_diag:0717_033332: # pmap -r 519 ...
sys_diag:0717_033332: # ptree -a 519 ...
sys_diag:0717_033332: # pfiles 519 ...
sys_diag:0717_033333: Dtrace: IO by process 519 ... (_dpio Snap 0)
sys_diag:0717_033339: # pmap -xs 448 ...
sys_diag:0717_033339: # pmap -S 448 ...
sys_diag:0717_033339: # pmap -r 448 ...
sys_diag:0717_033339: # ptree -a 448 ...
sys_diag:0717_033339: # pfiles 448 ...
sys_diag:0717_033340: Dtrace: IO by process 448 ... (_dpio Snap 0)
sys_diag:0717_033346: # pmap -xs 90 ...
sys_diag:0717_033346: # pmap -S 90 ...
sys_diag:0717_033346: # pmap -r 90 ...
sys_diag:0717_033346: # ptree -a 90 ...
sys_diag:0717_033346: # pfiles 90 ...
sys_diag:0717_033347: Dtrace: IO by process 90 ... (_dpio Snap 0)
sys_diag:0717_033353: # pmap -xs 825 ...
sys_diag:0717_033353: # pmap -S 825 ...
sys_diag:0717_033353: # pmap -r 825 ...
sys_diag:0717_033353: # ptree -a 825 ...
sys_diag:0717_033353: # pfiles 825 ...
sys_diag:0717_033353: Dtrace: IO by process 825 ... (_dpio Snap 0)
sys_diag:0717_033353: # /usr/bin/netstat -i -a ...
sys_diag:0717_033400: # Snapshot Kernel Memory Usage.. ::memstat | mdb -k ...
sys_diag:0717_033409: # /usr/sbin/lockstat -IW -n 100000 -s 13 sleep 5 ...
sys_diag:0717_033419: # /usr/sbin/lockstat -A -n 90000 -D15 sleep 5 ...
sys_diag:0717_033431: # /usr/sbin/lockstat -A -s8 -n 90000 -D10 sleep 5 ...
sys_diag:0717_033446: # /usr/sbin/lockstat -AP -n 90000 -D10 sleep 5 ...
sys_diag:0717_033521: Dtrace: Involuntary Context Switches (icsw) by process .. (_dmpc Snap 0)
sys_diag:0717_033526: Dtrace: Cross CPU Calls (xcal) caused by process ........ (_dmpc Snap 0)
sys_diag:0717_033531: Dtrace: MUTEX try lock (smtx) by lwp/process ............ (_dmpc Snap 0)

sys_diag: --\*\*-- (Background) DATA COLLECTION FOR 298 secs STARTED --\*\*--
sys_diag:0717_033531: # /usr/bin/vmstat -q 1 298 > ./sysd_socrates_070717_0332/sysd_vm_socrates_070717_033209.out 2>&1 &
sys_diag:0717_033531: # /usr/bin/iostat -xn 1 298 > ./sysd_socrates_070717_0332/sysd_io_socrates_070717_033209.out 2>&1 &
sys_diag:0717_033531: # /usr/bin/mpstat -q 1 298 > ./sysd_socrates_070717_0332/sysd_mp_socrates_070717_033209.out 2>&1 &
sys_diag:0717_033537: # /usr/bin/netstat -i -I lo0 1 298 > ./sysd_socrates_070717_0332/sysd_net1_socrates_070717_033537.out 2>&1 &
sys_diag:0717_033537: # /usr/bin/kstat -p -T u -n lo0 1> ./sysd_socrates_070717_0332/sysd_knetb_lo0_socrates_070717_033537.out 2>&1
sys_diag:0717_033538: # /usr/bin/netstat -i -I hme0 1 298 > ./sysd_socrates_070717_0332/sysd_net2_socrates_070717_033538.out 2>&1 &
sys_diag:0717_033538: # /usr/bin/kstat -p -T u -n hme0 1> ./sysd_socrates_070717_0332/sysd_knetb_hme0_socrates_070717_033538.out 2>&1
sys_diag:0717_033538: # /usr/sbin/snoop ...

sys_diag: ------- (Foreground) Gathering System Configuration Details -------
sys_diag:0717_033539: # uname -a ...
sys_diag:0717_033539: # hostid ...
sys_diag:0717_033539: # domainname (DNS) ...
sys_diag:0717_033539: ###### SYSTEM CONFIGURATION / DEVICE INFO ######
sys_diag:0717_033539: # prtdiag ...
sys_diag:0717_033539: # prtconf | grep Memory ...
sys_diag:0717_033539: # /usr/sbin/psrinfo -v ...
sys_diag:0717_033539: # /usr/sbin/psrinfo -pv ...
sys_diag:0717_033539: # /usr/sbin/psrset -q ...
sys_diag:0717_033539: # cfgadm -l ...
sys_diag:0717_033539: # cfgadm -al ...
sys_diag:0717_033539: # cfgadm -v ...
sys_diag:0717_033539: # cfgadm -av | grep memory | grep perm ...
sys_diag:0717_033541: ###### E10K / E25K / SunFire System INFO ######
sys_diag:0717_033541: # Checking Kernel Cage settings ...
sys_diag:0717_033541: # eeprom ...
sys_diag:0717_033541: # /usr/bin/coreadm ...
sys_diag:0717_033541: # /usr/sbin/dumpadm ...
sys_diag:0717_033541: # modinfo ...
sys_diag:0717_033541: # /usr/sbin/lustatus ...
sys_diag:0717_033541: # cat /etc/path_to_inst ...
sys_diag:0717_033542: ###### WORKLOAD CHARACTERIZATION ######
sys_diag:0717_033542: # prstat -c -a 1 1 ...
sys_diag:0717_033542: # prstat -c -J 1 1 ...
sys_diag:0717_033542: # prstat -c -Z 1 1 ...
sys_diag:0717_033542: # prstat -c 1 2 ...
sys_diag:0717_033544: # prstat -c -v 1 3 ...
sys_diag:0717_033546: # ps -e -o ...(by %CPU) ...
sys_diag:0717_033546: # ps -e -o ...(by %MEM) ...
sys_diag:0717_033546: # ps -e -o ...(by LWP) ...
sys_diag:0717_033546: ###### PERFORMANCE PROFILING (System / Kernel) ######
sys_diag:0717_033547: # vmstat 1 5 ...
sys_diag:0717_033551: # /usr/bin/mpstat 1 3 ...
sys_diag:0717_033551: # /usr/bin/isainfo -v ...
sys_diag:0717_033553: # /usr/bin/ipcs -a ...
sys_diag:0717_033553: # /usr/bin/pagesize ...
sys_diag:0717_033553: # swap -l ...
sys_diag:0717_033553: # swap -s ...
sys_diag:0717_033553: # /usr/bin/vmstat -s ...
sys_diag:0717_033553: # /usr/bin/kstat -n system_pages ...
sys_diag:0717_033553: # /usr/bin/kstat -n vm ...
sys_diag:0717_033554: # /usr/sbin/trapstat 1 2 ...
sys_diag:0717_033554: # /usr/sbin/trapstat -t 1 2 ...
sys_diag:0717_033554: # /usr/sbin/trapstat -l ...
sys_diag:0717_033554: # /usr/sbin/trapstat -t 1 2 ...
sys_diag:0717_033554: # /usr/sbin/trapstat -T 1 2 ...
sys_diag:0717_033554: # /usr/sbin/intrstat 1 2 ...
sys_diag:0717_033554: # /usr/bin/vmstat -i ...
sys_diag:0717_033554: ###### KERNEL ZONES/ SRM / Acctg / TUNABLES ######
sys_diag:0717_033554: # /usr/sbin/zoneadm list -v ...
sys_diag:0717_033554: # /usr/bin/projects -l ...
sys_diag:0717_033554: # /usr/sbin/psrset -i ...
sys_diag:0717_033554: # /usr/sbin/psrset -p ...
sys_diag:0717_033554: # /usr/sbin/psrset -q ...
sys_diag:0717_033554: # /usr/sbin/rctladm -l ...
sys_diag:0717_033554: # /usr/bin/priocntl -l ...
sys_diag:0717_033554: # /usr/sbin/acctadm ...
sys_diag:0717_033554: # /usr/sbin/acctadm -r...
sys_diag:0717_033554: # tail -80 /etc/system ...
sys_diag:0717_033554: # sysdef | tail -85 ...
sys_diag:0717_033554: # tail -40 /etc/init.d/sysetup ...
sys_diag:0717_033554: # cat /etc/power.conf ...
sys_diag:0717_033612: ###### STORAGE / ARRAY INFO ######
sys_diag:0717_033612: # prtconf -pv ...
sys_diag:0717_033613: # luxadm probe ...
sys_diag:0717_033614: ###### STORAGE VOLUME MANAGEMENT INFO ######
sys_diag:0717_033614: ###### SOLARIS (SDS/SVM) VOLUME MANAGER Info ######
sys_diag:0717_033614: # /sbin/metadb ...
sys_diag:0717_033614: # /sbin/metastat ...
sys_diag:0717_033614: # /sbin/metastat -p...
sys_diag:0717_033614: ###### Sun STMS / MPxIO Info ######
sys_diag:0717_033614: # cat /kernel/drv/fp.conf ...
sys_diag:0717_033614: # cat /kernel/drv/fcp.conf ...
sys_diag:0717_033614: ###### FILESYSTEM INFO ######
sys_diag:0717_033614: # df ...
sys_diag:0717_033614: # df -k ...
sys_diag:0717_033614: # mount -v ...
sys_diag:0717_033614: # /usr/sbin/showmount -a ...
sys_diag:0717_033614: # cat /etc/vfstab ...
sys_diag:0717_033614: # /usr/bin/cachefsstat ...
sys_diag:0717_033614: ###### I/O STATS ######
sys_diag:0717_033614: # /usr/bin/iostat -nxe 3 2 ...
sys_diag:0717_033614: # /usr/bin/iostat -xcC 3 2 ...
sys_diag:0717_033614: # /usr/bin/iostat -xnE ...
sys_diag:0717_033614: ###### NFS INFO ######
sys_diag:0717_033614: # /usr/bin/nfsstat ...
sys_diag:0717_033614: # /usr/bin/nfsstat -m ...
sys_diag:0717_033614: ###### NETWORKING INFO ######
sys_diag:0717_033614: # cat /etc/hosts ...
sys_diag:0717_033614: # /usr/sbin/ifconfig -a ...
sys_diag:0717_033614: # /usr/bin/netstat -i ...
sys_diag:0717_033614: # /usr/bin/netstat -r ...
sys_diag:0717_033614: # /usr/sbin/arp -a ...
sys_diag:0717_033614: # /usr/sbin/ping -s 56 10 ...
sys_diag:0717_033614: # /usr/sbin/ping -s 1016 10 ...
sys_diag:0717_033614: # /usr/sbin/ping -s 56 10 ...
sys_diag:0717_033614: # /usr/sbin/ping -s 1016 10 ...
sys_diag:0717_033614: # cat /etc/hostname.hme0 ...
sys_diag:0717_033614: # cat /etc/inet/networks ...
sys_diag:0717_033614: # cat /etc/netmasks ...
sys_diag:0717_033614: # tail -30 /etc/inet/ntp.server ...
sys_diag:0717_033614: # /usr/sbin/dladm show-dev ...
sys_diag:0717_033614: # /usr/sbin/dladm show-link ...
sys_diag:0717_033614: # /usr/sbin/dladm show-aggr ...
sys_diag:0717_033614: # /usr/sbin/pntadm -L ...
sys_diag:0717_033703: # /usr/bin/kstat -c net ...
sys_diag:0717_033703: # ndd -get /dev/tcp ...
sys_diag:0717_033703: # ndd -get /dev/udp ...
sys_diag:0717_033703: # ndd -get /dev/ip ...
sys_diag:0717_033706: # ndd -set /dev/hme instance 0 ...
sys_diag:0717_033706: # ndd -get /dev/hme ...
sys_diag:0717_033706: # /usr/bin/netstat -a ...
sys_diag:0717_033711: # /usr/bin/netstat -s ...
sys_diag:0717_033711: ###### TTY / MODEM INFO ######
sys_diag:0717_033711: # /usr/sbin/pmadm -l ...
sys_diag:0717_033711: # cat /etc/remote ...
sys_diag:0717_033711: # cat /var/adm/aculog ...
sys_diag:0717_033711: ###### USER / ACCOUNT / GROUP Info ######
sys_diag:0717_033711: # w ...
sys_diag:0717_033711: # who -a ...
sys_diag:0717_033711: # cat /etc/passwd ...
sys_diag:0717_033711: # cat /etc/group ...
sys_diag:0717_033711: ###### SERVICES / NAMING RESOLUTION ######
sys_diag:0717_033711: # /usr/bin/svcs -v ...
sys_diag:0717_033711: # cat /etc/services ...
sys_diag:0717_033711: # cat /etc/inetd.conf ...
sys_diag:0717_033711: # cat /etc/inittab ...
sys_diag:0717_033711: # cat /etc/nsswitch.conf ...
sys_diag:0717_033711: # cat /etc/resolv.conf ...
sys_diag:0717_033711: # cat /etc/auto_master ...
sys_diag:0717_033711: # cat /etc/auto_home ...
sys_diag:0717_033712: # /usr/bin/ypwhich ...
sys_diag:0717_033712: # /usr/bin/nisdefaults ...
sys_diag:0717_033712: ###### SECURITY / CONFIG FILES ######
sys_diag:0717_033712: # cat /etc/syslog.conf ...
sys_diag:0717_033712: # cat /etc/pam.conf ...
sys_diag:0717_033712: # cat /etc/default/login ...
sys_diag:0717_033712: # tail -250 /var/adm/sulog ...
sys_diag:0717_033712: # /usr/bin/last reboot ...
sys_diag:0717_033712: # /usr/bin/last -200 ...
sys_diag:0717_033712: # /usr/sbin/ipf -T list ...
sys_diag:0717_033712: # cat /etc/ipf/ipf.conf ...
sys_diag:0717_033712: # cat /etc/ipf/pfil.ap ...
sys_diag:0717_033712: # /usr/sbin/ipnat -vls ...
sys_diag:0717_033713: ###### HA/ CLUSTERING INFO ######
sys_diag:0717_033713: ###### SUN N1 Configuration INFO ######
sys_diag:0717_033713: ###### APPLICATION / ORACLE CONFIG FILES ######
sys_diag:0717_033713: ###### PACKAGE INFO / SOLARIS REGISTRY ######
sys_diag:0717_033713: # /usr/bin/prodreg browse ...
sys_diag:0717_033713: # /usr/bin/pkginfo ...
sys_diag:0717_033713: # /usr/bin/pkginfo -l ...
sys_diag:0717_033713: ###### PATCH INFO ######
sys_diag:0717_033713: # /usr/bin/showrev -p ...
sys_diag:0717_033713: # /usr/sadm/bin/smpatch analyze NOT RUN, passwd required....
sys_diag:0717_033753: ###### CRONTAB FILE LISTINGS ######
sys_diag:0717_033753: ###### FMD / SYSTEM MESSAGE/LOG FILES ######
sys_diag:0717_033753: # /usr/sbin/fmadm config ...
sys_diag:0717_033753: # /usr/sbin/fmdump ...
sys_diag:0717_033753: # /usr/sbin/fmstat ...
sys_diag:0717_033753: # tail -250 /var/adm/messages ...
sys_diag:0717_033753: # /usr/bin/dmesg ...
sys_diag:0717_033753: # tail -500 /var/log/syslog ...
sys_diag:0717_033754: ...WAITING 12 seconds for midpoint data collection...

sys_diag: ------- MidPoint Process SNAPSHOT (# 1) -------
sys_diag:0717_033806: Dtrace: TCP write bytes by process ...(_dtcp_tx Snap 1)
sys_diag:0717_033806: Dtrace: TCP read bytes by process ... (_dtcp_rx Snap 1)
sys_diag:0717_033806: Dtrace: systemwide IO / IO wait... (_diow Snap 1)
sys_diag:0717_033832: Dtrace: Syscall count by process... (_dcalls_ Snap 1)
sys_diag:0717_033840: Dtrace: Syscall count by syscall... (_dsyscall_ Snap 1)
sys_diag:0717_033847: Dtrace: Read bytes by process... (_dR_ Snap 1)
sys_diag:0717_033855: Dtrace: Write bytes by process... (_dW_ Snap 1)
sys_diag:0717_033903: Dtrace: Sysinfo counts by process... (_dsinfo_ Snap 1)
sys_diag:0717_033911: Dtrace: Sdt_counts ... (_dsdtcnt_ Snap 1)
sys_diag:0717_033918: Dtrace: Interupt Times [sdt:::intr].. (_dintrtm_ Snap 1)
sys_diag:0717_033918: # ps -e -o ...(by %CPU) ... Snapshot # 1
sys_diag:0717_033918: # ps -e -o ...(by %MEM) ... Snapshot # 1
sys_diag:0717_033929: # pmap -xs 4188 ...
sys_diag:0717_033929: # pmap -S 4188 ...
sys_diag:0717_033929: # pmap -r 4188 ...
sys_diag:0717_033929: # ptree -a 4188 ...
sys_diag:0717_033929: # pfiles 4188 ...
sys_diag:0717_033929: Dtrace: IO by process 4188 ... (_dpio Snap 1)
sys_diag:0717_033935: # pmap -xs 4181 ...
sys_diag:0717_033935: # pmap -S 4181 ...
sys_diag:0717_033935: # pmap -r 4181 ...
sys_diag:0717_033935: # ptree -a 4181 ...
sys_diag:0717_033935: # pfiles 4181 ...
sys_diag:0717_033936: Dtrace: IO by process 4181 ... (_dpio Snap 1)
sys_diag:0717_033942: # /usr/bin/netstat -i -a ...
sys_diag:0717_033942: # Snapshot Kernel Memory Usage.. ::memstat | mdb -k ...
sys_diag:0717_033952: # /usr/sbin/lockstat -IW -n 100000 -s 13 sleep 5 ...
sys_diag:0717_034002: # /usr/sbin/lockstat -A -n 90000 -D15 sleep 5 ...
sys_diag:0717_034015: # /usr/sbin/lockstat -A -s8 -n 90000 -D10 sleep 5 ...
sys_diag:0717_034037: # /usr/sbin/lockstat -AP -n 90000 -D10 sleep 5 ...
sys_diag:0717_034051: Dtrace: Involuntary Context Switches (icsw) by process .. (_dmpc Snap 1)
sys_diag:0717_034056: Dtrace: Cross CPU Calls (xcal) caused by process ........ (_dmpc Snap 1)
sys_diag:0717_034101: Dtrace: MUTEX try lock (smtx) by lwp/process ............ (_dmpc Snap 1)

sys_diag: ------- EndPoint Process SNAPSHOT (# 2) -------
sys_diag:0717_034101: # /usr/bin/kstat -p -T u -n lo0 2>&1
sys_diag:0717_034101: # /usr/bin/kstat -p -T u -n hme0 2>&1
sys_diag:0717_034107: Dtrace: TCP write bytes by process ...(_dtcp_tx Snap 2)
sys_diag:0717_034107: Dtrace: TCP read bytes by process ... (_dtcp_rx Snap 2)
sys_diag:0717_034107: Dtrace: systemwide IO / IO wait... (_diow Snap 2)
sys_diag:0717_034133: Dtrace: Syscall count by process... (_dcalls_ Snap 2)
sys_diag:0717_034141: Dtrace: Syscall count by syscall... (_dsyscall_ Snap 2)
sys_diag:0717_034149: Dtrace: Read bytes by process... (_dR_ Snap 2)
sys_diag:0717_034156: Dtrace: Write bytes by process... (_dW_ Snap 2)
sys_diag:0717_034204: Dtrace: Sysinfo counts by process... (_dsinfo_ Snap 2)
sys_diag:0717_034212: Dtrace: Sdt_counts ... (_dsdtcnt_ Snap 2)
sys_diag:0717_034220: Dtrace: Interupt Times [sdt:::intr].. (_dintrtm_ Snap 2)
sys_diag:0717_034220: # ps -e -o ...(by %CPU) ... Snapshot # 2
sys_diag:0717_034220: # ps -e -o ...(by %MEM) ... Snapshot # 2
sys_diag:0717_034230: # pmap -xs 519 ...
sys_diag:0717_034230: # pmap -S 519 ...
sys_diag:0717_034230: # pmap -r 519 ...
sys_diag:0717_034230: # ptree -a 519 ...
sys_diag:0717_034230: # pfiles 519 ...
sys_diag:0717_034231: Dtrace: IO by process 519 ... (_dpio Snap 2)
sys_diag:0717_034237: # pmap -xs 448 ...
sys_diag:0717_034237: # pmap -S 448 ...
sys_diag:0717_034237: # pmap -r 448 ...
sys_diag:0717_034237: # ptree -a 448 ...
sys_diag:0717_034237: # pfiles 448 ...
sys_diag:0717_034238: Dtrace: IO by process 448 ... (_dpio Snap 2)
sys_diag:0717_034244: # pmap -xs 90 ...
sys_diag:0717_034244: # pmap -S 90 ...
sys_diag:0717_034244: # pmap -r 90 ...
sys_diag:0717_034244: # ptree -a 90 ...
sys_diag:0717_034244: # pfiles 90 ...
sys_diag:0717_034245: Dtrace: IO by process 90 ... (_dpio Snap 2)
sys_diag:0717_034251: # pmap -xs 825 ...
sys_diag:0717_034251: # pmap -S 825 ...
sys_diag:0717_034251: # pmap -r 825 ...
sys_diag:0717_034251: # ptree -a 825 ...
sys_diag:0717_034251: # pfiles 825 ...
sys_diag:0717_034251: Dtrace: IO by process 825 ... (_dpio Snap 2)
sys_diag:0717_034251: # /usr/bin/netstat -i -a ...
sys_diag:0717_034258: # Snapshot Kernel Memory Usage.. ::memstat | mdb -k ...
sys_diag:0717_034307: # /usr/sbin/lockstat -IW -n 100000 -s 13 sleep 5 ...
sys_diag:0717_034317: # /usr/sbin/lockstat -A -n 90000 -D15 sleep 5 ...
sys_diag:0717_034329: # /usr/sbin/lockstat -A -s8 -n 90000 -D10 sleep 5 ...
sys_diag:0717_034344: # /usr/sbin/lockstat -AP -n 90000 -D10 sleep 5 ...
sys_diag:0717_034358: Dtrace: Involuntary Context Switches (icsw) by process .. (_dmpc Snap 2)
sys_diag:0717_034404: Dtrace: Cross CPU Calls (xcal) caused by process ........ (_dmpc Snap 2)
sys_diag:0717_034408: Dtrace: MUTEX try lock (smtx) by lwp/process ............ (_dmpc Snap 2)

sys_diag:0717_034408: ------- Data Collection COMPLETE -------
sys_diag:0717_034408: ###### SYSTEM ANALYSIS : INITIAL FINDINGS ... ######
sys_diag:0717_034414: ###### PERFORMANCE DATA : POTENTIAL ISSUES ######

sys_diag:0717_034414: ## Analyzing VMSTAT CPU Datafile :
	./sysd_socrates_070717_0332/sysd_vm_socrates_070717_033209.out ...

\* NOTE: 2.6936 % : 8 of 297 VMSTAT CPU entries are WARNINGS!! \*

TOTAL CPU AVGS : RUNQ= 0.1 : BThr= 0.0 : USR= 15.0 : SYS= 11.2 : IDLE= 73.5
PEAK CPU HWMs : RUNQ= 8 : BThr= 0 : USR= 51 : SYS= 96 : IDLE= 0


sys_diag:0717_034414: ## Analyzing VMSTAT MEMORY from Datafile :
./sysd_socrates_070717_0332/sysd_vm_socrates_070717_033209.out ...

\* NOTE: 0.673401 % : 2 of 297 VMSTAT MEMORY entries are WARNINGS!! \*

TOTAL MEM AVGS : SR= 0.0 : SWAP_free= 747697.4 K : FREE_RAM= 287786.6 K
PEAK MEM Usage: SR= 0 : SWAP_free= 500128.0 K : FREE_RAM= 57080.0 K


sys_diag:0717_034414: ## Analyzing MPSTAT Datafile : ./sysd_socrates_070717_0332/sysd_mp_\*.out ...

\* NOTE: 5.20134 % : 31 of 596 MPSTAT CPU entries are WARNINGS!! \*

CPU MP AVGS: Wt= 0: Xcal= 736: csw= 120: icsw= 3: migr= 5: smtx= 3: syscl= 1024
PEAK MP HWMs: Wt= 0: Xcal= 51771: csw= 14108: icsw= 32: migr= 55: smtx= 79: syscl= 25836

NOTE: 0.2% CPU cycles handling TLB MISSES (0.0% ITLB_misses: 0.2% DTLB_misses)


sys_diag:0717_034414: ## Analyzing IOSTAT Datafile :
./sysd_socrates_070717_0332/sysd_io_\*.out ...

\* NOTE: 14.4578 % : 24 of 166 IOSTAT entries are WARNINGS!! \*

TOP 10 Slowest IO Devices (\* AVG of non-zero device entries \*) :

r/s w/s kr/s kw/s actv wsvc_t asvc_t %w %b device # I/O Samples

32.6 10.8 263.6 24.6 0.8 0.0 13.7 0.0 19 c0t0d0 164
34.0 7.5 10.8 0.0 0.0 0.0 0.0 0.0 0 c0t1d0 2


CONTROLLER IO : AVG and TOTAL Throughput per HBA (\*active/non-zero entries only\*) :

c0 : AVG : 32.6 r/s | 10.8 w/s | 260.6 kr/s | 24.3 kw/s |
c0 : TOTAL: 5408 r | 1790 w | 43258 kr | 4037 kw | 166 entries


sys_diag:0717_034414: ## Analyzing NETSTAT Datafiles : ...

\* lo0 : NOTE: 0 % : 0 of 297 NETSTAT entries are WARNINGS!! \*
\* hme0 : NOTE: 0 % : 0 of 297 NETSTAT entries are WARNINGS!! \*

NET1 : lo0 : 4 0.0 0.0 0.0 0.0 0.0

NET2 : hme0 : 14 0.4 0.0 0.4 0.0 0.0
: hme0 : TOT_RX_Bytes TOT_TX_Bytes TOT_RX_Packets TOT_TX_Packets TOTAL_Seconds
22210 30348 124 112 328
: hme0:1: TOT_RX_Packets TOT_TX_Packets

: hme0:1: 0 0

NOTE: \*\* 2 ESTABLISHED connections (sockets) exist\*\*




sys_diag:0717_034417: ... gen_html_hdr ...
sys_diag:0717_034417: ... gen_html_rpt ...

sys_diag:0717_034419: ## Generating TAR file : ./sysd_socrates_070717_0332.tar ...

tar -cvf ./sysd_socrates_070717_0332.tar ./sysd_socrates_070717_0332 1>/dev/null
compress ./sysd_socrates_070717_0332.tarData files have been TARed and compressed in :

\*\*\* ./sysd_socrates_070717_0332.tar.Z \*\*\*

------- Sys_Diag Complete -------

( Copyright 2007, Todd A. Jobson )
Add to Technorati Favorites

Thursday Jul 12, 2007

What is sys_diag ?.. Automating Solaris Performance Profiling and Workload Characterization.

The following is an excerpt from the README_sys_diag.txt file which gives a
high level overview of the sys_diag capabilities and command line arguments /
/ usage examples... I've created this over many years to automate and reduce
the amount of time it takes to gather and correlate system data for conducting off-site (remote) Performance and Configuration Analysis.  With sys_diag
all you need to do is download the ksh script.. and you're on your way. After
it's run.. you get a single .tar.Z that you can upload or email for remote
analysis.. (-G even includes a wide variety of DTrace examination..).

Read the following introduction.. and more specific examples will follow.


sys_diag v.7.04 Overview :


  sys_diag is a Solaris utility (ksh script) that can perform several
  functions, among them, system configuration 'snapshot' and reporting
  (detailed or high-level) plus workload characterization/profiling via performance
  data gathering (over some specified duration or time in point 'snapshot'),
  high-level analysis, and reporting of findings/exceptions (based upon
  perf thresholds that can be easily changed within the script header).

  The output is provided in a single .tar.Z of output and corresponding
  data files, and a local sub-directory where report/data files are stored.
  The report format is provided in .html, .txt, and .ps as a single file
  for easy review (without requiring trudging through several subdirectories
  of separate files to manually correlate and review).

  sys_diag runs on any Solaris 2.6 (or above) Sun platform, including
  reporting of new Solaris 10 capabilities (zone/containers, SVM,
  zfspools, fmd, ipfilter/ipnat, link aggr, Dtrace probing, etc...).

  Beyond the Sun configuration reporting commands [System/storage HW config,
  OS config, kernel tunables, network/IPMP/Trunking/LLT config, FS/VM/NFS,
  users/groups, security, NameSvcs, pkgs, patches, errors/warnings, and
  system/network performance metrics...], sys_diag also captures relevant
  application configuration details, such as Sun N1, Sun Cluster 2.x/3.x,
  Veritas VCS/VM/vxfs.., Oracle .ora/listener files, etc.. detailed
  configuration capture of key files (and tracking of changes via -t), etc ...

  Of all the capabilities, the greatest benefits are found by being able
  to run this single ksh script on a system and do the analysis  from one single report/
  file... offline/elsewhere (in addition to being capable of  historically
  archiving system configurations, for disaster recovery.. or to allow for
  tracking system chgs over time.. after things are built/tested/certified).

  One nice feature for performance analysis is that the vmstat and netstat
  data is exported in a text format friendly to import and created graphs
  from in StarOffice or Excell.. as well as creating IO and NET device
  Averages from IOSTAT / Netstat data (# IO's per device, AVG R/W K, etc..)
  along with peak exceptions for CPU / MEM / IO / NET ..

  Although this tool isn't meant to replace long-term historical Performance
  Trending and Capacity Planning packages (Teamquest, etc..), it provides the
  foundation and basis for a very robust starting point (and actually is much
  better at point in time workload characterization and root cause analysis of
  bottlenecks, where very granular detailed data correlation is required).

  Even though I'm a Sun employee, this has been personally developed over many
  years, in my spare time in order to make my life a lot easier and
  more efficient.  Hopefully others will find this utility capable of
  doing the same for them, also making use of it's legwork.. to streamline
  the admin/analysis activities required of them.  This has been an invaluable
  tool used to diagnose / analyze hundreds of performance and/or configs issues

  Regarding the system overhead, sys_diag runs all commands in a serial
  fashion (waiting for each command to complete before running the next)
  impacting system performance the same as if an admin were typing these
  commands one at a time on a console.. with the exception of the background
  vmstat/mpstat/iostat/netstat that's done when gathering performance data
  (-g | -G) over some interval for report/analysis (which generally has minimal
  impact on a system, especially if the sample interval [-I] is not every
  second, or if the lighter weight -g is run vs. -G detailed/Dtrace snapshots).

  sys_diag is generally run from /var/tmp as "sys_diag -l"  for creating
  a detailed long report, or via "sys_diag -g -l " for gathering
  performance data and generating a long/detailed config/analysis report),
  however offers many command line parameters documented within the header,
  or via "sysdiag -?".   \*\* READ the Usage below, as well as the Performance
  Parameters sections for further enlightenment.. ;)

  NOTE: For the best .html viewing experience, Do NOT use MS Internet Explorer browser
        as it varies in support of HTML stds for formating and iframe file inclusion
        (ending up opening many windows vs embedding output files within
        the single .html report).  \*\* USE Netscape, Mozilla, Firefox, etc.. browsers,
        ensuring that your display resolution is set to the maximum resolution, and
        font sizes are defaults or not made too large (for best viewing open full screen)

    \*\*\* As is the best practice for any environment, first TEST thoroughly on a representative
        TEST configuraiton PRIOR to running this or making any production system changes.
        (read the sys_diag ksh headers for disclaimer and support notes) \*\*\*

\*\* See  :  for other entries relating to system performance,
                capacity planning, and systems architecture / availability.

\*\* For the latest release of sys_diag see either BigAdmin or at the following URL's : 



Common Command Line usage and available parameters :


	# sys_diag [-a -A -c -C -d_ -D -f_ -g -G -H -I_ -l -L_ -n -o_ -p -P -s -S -T_ -t -u -v -V -h|-? ]

-a Application details (included in -l/-A)
-A ALL Options are turned on, except Debug and -u
-c Configuration details (included in -l/-A)
-C Cleanup Files and remove Directory if tar works
-d path Base directory for data directory / files
-D Debug mode (ksh set -x .. echo statements/variables/evaluations)
-f input_file Used with -t to list configuration files to Track changes for
-g gather Performance data (2 sec intervals for 5 mins, unless -I |-T exist)
-G GATHER Extra Perf data (S10 Dtrace, more lockstats, pmap/pfiles) vs -g
-h | -? Help / Command Usage (this listing)
-H HA config and stats
-I secs Perf Gathering Sample Interval (default is 2 secs)
-l Long Listing (most details, but not -g,-V,-A,-t,-D)
-L label_descr_nospaces (Descriptive Label For Report)
-n Network configuration and stats (also included in -l/-A except ndd settings)
-o outfile Output filename (stored under sub-dir created)
-p Generate Postscript Report, along with .txt, and .html
-P -d ./data_dir_path Post-process the Perf data skipped with -S and finish .html rpt
-s SecurIty configuration
-S SKIP POST PROCESSing of Performance data (use -P -d data_dir to complete)
-t Track configuration / cfg_file changes (Saves/Rpts cfg/file chgs \*see -f)
-T secs Perf Gathering Total Duration (default is 300 secs =5 mins)
-u unTar ed: (do NOT create a tar file)
-v version Information for sys_diag
-V Verbose Mode (adds path_to_inst, network dev's ndd settings, mdb, snoop..)
Longer message/error/log listings. Additionally, pmap is run if -g ||-G,
and the probe duration for Dtrace and lockstat sampling is widened
from 2 seconds (during -G) to 5 seconds (if -G && -V). Ping is
also run against the default route and to guage latency.

NOTE: NO args equates to a brief rpt (No -A,-g/I,-l,-t,-D,-V,..)

\*\* Also, note that option/parameter ordering is flexible, as well as use of white
space before arguments to parameters (or not). The only requirement is to list
every option/parameter separately with a preceeding - (-g -l , but not -gl).

BOTH of the following command line syntax examples is functionally the same :

eg. ./sys_diag -g -I 1 -T 1800 -t -f ./config_files -l
./sys_diag -g -l -t -f./config_files -I1 -T1800



  eg.   Common Usage :
./sys_diag -l Creates a LONG /detailed configuration rpt (.html/.txt)
Without -l, the config report created has basic system cfg details.

./sys_diag -g -l gathers performance data at the default sampling rate of 2 secs for

a total duration of 5 mins, adding a color coded performance header/
Dashboard Summary section and any performance findings/
exceptions found to the long (-l) cfg rpt. Also takes (3) starting/
midpt/endpoint snapshots using minimal lockstat/kstat (1sec)
			     NOTE: -g is meant to gather perf data without overhead, therefore
  only 1 second lockstat samples are taken. Use -G and/or -V
for more detailed system probing (see examples and notes below)
Using -V with -g, adds pmap/pfiles snapshots, vs. using -G
to also capture Dtrace and extended lockstat probing.
                           \* Any time that sys_diag is run with either -g or -G, the command
line output is appended to the file sys_diag_perflog.out, which
gets copied and archived as part of the final .tar.Z output file.

  ./sys_diag -g -I 1 -T 600 -l gathers perf data at 1 sec samples for 10 mins and
creates a long config rpt as noted above. Also does

basic start/mid/endpoint sampling using lockstat/kstat/pmap.

./sys_diag -l -C creates long config rpt, and Cleans up..

aka removes the data directory after tar.Z completes

./sys_diag -d base_directory_path (changes the base dir for datafiles from curr dir)

./sys_diag -G -I 1 -T 600 -l Gathers DEEP performance & Dtrace/lockstat/pmap data

at 1 sec samples for 10 mins & creates a long cfg rpt
(in addition to the standard data gathering from -g).

\*NOTE: this runs all Dtrace/Lockstat/Pmap probing during 3 snapshot intervals

(beginning_0/midpoint_1/ and endpoint_#2 snapshots), limiting probing
overhead to BEFORE/AFTER the standard data gathering begins
(vmstat, mpstat, iostat, netstat, .. from -g).
The MIDPOINT probing occurs at a known point as not to confuse this
activity for other system processing.

\*Because of this, standard data collection may not start for 30+ seconds,
or until the beginning snapshot (snapshot_#0) is complete.

                        (-g snapshot_#0 activities only take a couple seconds to complete, since 
they do not include any Dtrace/lockstat.. beyond 1 sec samples).

./sys_diag -G -V -I 1 -T 600 Gathers DEEP, VERBOSE, performance & Dtrace/lockstat/pmap

data at 1 sec samples for 10 mins (using 5 second Dtrace and
Lockstat snapshots, vs. 2 second probes for only -G.
(in addition to the standard data gathering from -g).

./sys_diag -g -l -S (gathers perf data, runs long config rpt, and SKIPS Post-Processing

and .html report generation)

\*\* This allows for completing the post-processing/analysis activities
either on another system, or at a later time, as long as the data_directory
exists (which can be extracted from the .tar.Z, then refered to as
-d data_dir_path ). \*\* See the next example using -P -d data_path \*\*

./sys_diag -P -d ./data_dir_path (Completes Skipped Post-Processing & .html rpt creation)

(Copyright 2007, Todd A. Jobson)
Add to Technorati Favorites

Tuesday Jul 10, 2007

Solaris Performance Analysis and Monitoring Tools... at what cost ?

In the area of Performance Analysis and related Monitoring tools, you'll find a plethora available for the Solaris environment. Each of them has it's own intrinsic costs associated.. listed here :

  • Monetary Costs ($$) :

    • Purchase Cost (Media, Documentation, etc..)
    • License Fees
    • Centralized or Management Server Required ? (HW Costs for System / Storage)
    • Hourly Costs of an Staff/ SME/ Consultant to Install/Config, Correlate, Interpret, Rpt findings...

  • Time / Effort Costs :

    • 3rd Party Installation / Configuration Pre-Requisites (libraries, tools Perl/Python.., etc..)
    • Server OS and Tools Design Requirements (Security, OS rev's, RAM, CPU, Storage, FS, Patches,..)
    • Server Installation (Rack/Stack, Network, Power, OS Install, Patching, Storage Cfg,..)
    • Server Toolset Installation (Installation, Configuration, License downloads, ..)
    • Client Node Agents Required for Installation / Configuration ?
    • Project/Manpower Time and Coordination Dependencies for an SME/Consultant vs. Other Resources (system, network, storage, etc...).
    • Time spent Installing, Configuring, Testing, Patching/Tweaking, Running, Correlating data, Analyzing/ Interpreting data, Reporting Findings...
    • Time spent learning the Toolset and how to interpret the raw and correlated data (thresholds, etc?)

  • System Overhead Costs :

    • CPU Consumption (% standard overhead vs. PEAK Load overhead)
    • Memory Consumption (RAM footprint.. standard vs. Peak load)
    • Storage Requirements (Toolset Installation space vs. active / historical storage vs. archiving req's)
    • RunTime Requirements (Running Constantly vs. During Specific PEAK load Intervals, Sampling Rate, ..)
    • Network Overhead (bandwidth and/or interrupt overhead due to data passed between client/server repository vs. local storage)
    • I/O Overhead (overhead of performing local IO.. generally depends on volume of data stored and sampling rates)

The Benefits of Accurate, Detailed, and Complete Data Gathering ...

\*\* NOTE: .. a key Attribute often left out is the ACCURACY and RELEVANCE of performance data captured (based up on the time it was captured, the sampling rates, and the level of detail provided).

This in many instances requires weighing the costs of having point in time event "detailed" snapshots (where the sampling rate intervals are very narrow.. per sec, etc.), vs. long-term historical trending data (where samples are aggregated and averaged over longer timeframes minimizing the storage requirements, but also smoothing out the Peak load visibility). For example, if you use a toolset or individual utility that can capture performance data at 1 second intervals, you will see a very granular view of systems utilization and PEAK load activity (resouce consumption, contention events, etc.).. VS.. using a historical trending toolset that can only save data at 1, 5, or 10 minute Averages.. (due to the contstraints of storage space available for the long periods of data that must be kept).

This might not seem like much would be missed, however.. even the difference between 1 second and 1 minute samples can be astronomical.. where 80 samples with 95% idle and 20 samples with 100% utilization (0% Idle) and a huge run queue will get "smoothed" out to a one minute sample where the box "appears" only 24% utilized (76% idle).. although the system is thrashing 20% of the time.  Even within the period of a second, you have over a billion instructions that get run on modern cpu's running at GHz + clock rates (Billions of cycles per second).. and only one aggregated sample for that period.

For complete end-to-end Capacity Planning and Performance Analysis capabilities, BOTH types of data is generally required (longer term trending for Capacity Planning purposes via graphs, etc.. VS. short term detailed drill down of system activity for point in time PEAK LOAD periods, allowing for detailed performance and utilization assessment / correlation).

Without detailed and granular data during peak periods, there can be no real correlation of root causes or specific bottlenecks... and in the same regard, without long-term, historical data that shows growth rates in capacity and cycles (patterns and models) of utilization and Peak activity.. accurate Capacity Planning isn't feasible.


..  if data captured doesn't include peak activity, or the granularity of samples is too sparse.. (not reflecting peak events), ...  then that data can only be useful for defining a BASELINE of Average Utilization.


MANY, many, .. tools


A wide variety of performance tools can be found.. from the high end.. using end-to-end third party products such as Teamquest (which provides a graphical, historical vantage point).. than need to be purchased, installed, and trained on... to the OS built-in utilities and the freely available open source / public domain variteies.

However, either way you go, be prepared for the requiring learning curve.. along with the extensive manual process and time required to identify and run the utilities, before you can capture and begin the extensive correlation process on the data from several disparate utilities (before you even get to do the analysis of your findings).

Either approach has it's advantages and disadvantages.. along with their strengths and weaknesses (3rd party purchased suites might save you time in graphical aggregation and correlation.. but tend to limit the level of detail and granularity available vs. what the OS utilities will provide).

The basic list of KEY "built-in" tools historically available for monitoring performance applies to nearly any Unix/Linux distribution, including the following partial list of common utilities used ... following the basic breakdown of computing subsystems :

\*\* CPU / Kernel Utilization :

--> vmstat (vm system cpu and kernel utilization metrics \*\* a great starting pt \*\*)
--> mpstat (multi processor .. per cpu performance statistics)

\*\* Memory / Kenel Utilization :

--> vmstat
--> ipcs

--> swap
--> top

\*\* I/O Performance :

--> iostat (Standard IO.. ufs, .. IO performance utility)
--> vxstat (Veritas vxfs filesystem IO performance)

\*\* Network Utilization :

--> netstat
--> ping
--> traceroute

\*\* Process / Kernel :

--> ps
--> top
--> prstat

--> ...

\* sar (provides most basic types of high level performance metrics, assuming that system accounting is turned on, which does incur some level of system overhead when always running)


SOLARIS 10 ... Above and Beyond other Unix / Linux Distributions ... 


In addition to the basic toolsets available, there exist the following key additions that Solaris 10 provides, which sets it apart from the other Unix / Linux variants.

\*\* DTrace (Dynamic Tracing via "D" language scripting and probe/providers)

__ Dtrace is the "Electron microscope" of performance analysis for a Solaris 10 system
See the DtraceToolkit for a long list of specific Dtrace scripts (several of which are used
within sys_diag, among others created)

\*\* lockstat (uses the kernel dtrace infrastructure) Summarizes system lock/mutex contention

\*\* Mdb (Modular Debugger)

\* kstat (Kernel statistics .. counters, etc..)

\* cpustat / cputrack (cpu statistics, system-wide or per process)

\* intrstat, trapstat (interrupt and system trap, I/DTLB_miss statistics, ..)

\* ... & many more.. [this list will be re-done in a future blog with a more thorough breakdown.. ]


The Time Saving.. automated nature... of SYS_DIAG   :)

Over the past several years, I have created a utility called "sys_diag" that offers the capability of automatically capturing performance statistics, using nearly all available system utilities.. and aggregating the data, performing analysis and HTML report generation of findings. Sys_diag creates a single .tar.Z compressed archive that can be emailed/ftp'd.. for performing system configuration and/or performance analysis off-site.. from virtually anywhere.. saving a LOT of time.. not requiring any 3rd party tools or agents to be installed on a system other than downloading the "sys_diag" ksh script itself (with a color coded dashboard.. and links to detailed analysis findings).  Virtually no learning curve is required for loading, running, and reflecting basic performance profiling, including high level subsystem bottlenecks (deeper root cause correlation might require some level of advanced sys admin knowledge).

Beyond performance analysis, sys_diag can be used to also generate a detailed configuration snapshot report, including OS, HW, Storage, SW, 3PP configuration attributes, among several other capabilities that it provides.

\*\* See the next blog entry for more details and examples on sys_diag \*\*.
The published repository and high level description of sys_diag is always available at BigAdmin using the following URL :

(Copyright 2007 Todd A. Jobson)

Add to Technorati Favorites

Wednesday May 16, 2007

What is Performance ? .. in the Real World

When we think of "Performance", the definition can have/take many connotations...

In the context of computing, the dictionary defines it as : (


"The manner in which or the efficiency with which something reacts or fulfills its intended purpose." or "the execution or accomplishment of work, acts, feats, etc."


From this definition, it can be readily seen that the "efficiency" and overall "utilization" of resources is a key characteristic of the "performance" of a system (also leaving room for some subjective interpretation).


Real World Performance.. and the holistic viewpoint

The other key aspects of assessing the performance, whether in the real world, or that of a system, relates directly to the volume of productive OUTPUT over a duration of TIME that a system produces.

In the arena of Information Technology.. as in real world performance (auto's, economics, the human body, etc..) the entity as a whole needs to be examined, allowing for symptoms to be identified in one or more areas... aka.. "sub-systems".   Hence, the complete Integrated "system" as a whole.. comes to life with it's own unique dynamics and patterns that need to be examined.

(eg.   one analogy might be the "performance" of a race car .. dependent upon the design/architecture of the vehicle.. and all it's constituent components.. the chasis [weight, flexibility, ..],  the steering [responsiveness, turn ratio..], the engine [horsepower, air/fule intake, exhaust output, ..], the Transmission [gear ratios, latency in shifting.., MTBF of clutch,..], braking [responsiveness, 60-0 secs, ..], tires [ G's on the skidpad, wear rate, ..], .. and Overall Performance.. [0-60 acceleration, MPG, top speed, slalom speed, ..] .. individually each can be measured easily.. but as a whole.. the INTEGRATED "system dynamics" come into play ). 

The same can be said (and analogous) to most "systems"... hence, looking at the environment as a whole is crucial ...


The "Application Environment" ... 

That holistic entity in the arena of Computing is called the "Application Environment".. comprised of all the systems and underlying nested/encapsulated sub-systems. In the IT world, an Application Environment is composed of all the underlying infrastructure that together provides and supports the "Service(s)" (environments, systems, networks, storage, OS, Application Software, etc...).


"Perceived" Performance and Expectations :

For any system (or environment), the ultimate guage is in the "Perception" of it's performance, relating to whether or not it can fullfil the expectations of it's client user community.

How efficient, proficient, and/or productive we perceive something to be, is in large part.. a product of our vantage point (perception), and how we judge or evaluate it... according to our expectations, pre-conceived notions (rules), and the means available to us for measuring it (tools, etc..).
The perception of one impatient user doesn't always accurately reflect the responsiveness, efficiency, or other attributes for evaluating the performance / workload characterization of a system.


Understanding , Metrics, and Measurement ...

From this vantage point, it becomes evident that in Assessing a system, there must be measurement of key attributes .. aka.. METRICS... and in order to define key metrics that can/should be monitored, we must first UNDERSTAND the system and how it works (components, mechanics, inputs / outputs, among other items that can be measured).

Hence, "If you can't understand it... you can't effectively measure it, .. and if you can't measure it.. you can't assess it..". (T.Jobson 7/2006)


Requirements dictate Measurements... driven by Service Level Commitments :

Of the various vantage points that a system's performance is guaged, the following
attributes (relating to specific Metrics that can be sampled) are typically those
which Service Level Agreements (SLA's) and/or Commitments (SLC's) are based upon (reflecting Customer Requirements and "acceptable" Thresholds.. ) :

  • Response Time (Client GUI's, Client/Server Transactions, Service Transactions,..) Measured as "acceptable" Latency.

  • Throughput (how much Volume of data can be pushed through a specific subsystem.. IO, Net..)
  • Transaction Rates (DataBase, Application Services, Infrastructure / OS / Network.. Services, etc.).  These can be either rates per Second, Hour, or even Day... measuring various service-related transactions.
  • Failure Rates (# or Frequency of exceeding  High or Low Water Marks .. aka Threshold Exceptions)
  • Resource Utilization (CPU Kernel vs. User vs. Idle, Memory Consumption, etc..)
  • Startup Time (System HW, OS boot, Volume Mgmt Mirroring, Filesystem validation, Cluster Data Services, etc..)
  • FailOver / Recovery Time (HA clustered DataServices, Disaster Recovery of a Geographic Service, ..)  Time to recover a failed Service (includes recovery and/or startup time of restoring the failed Service)

  • etc ...

    For any exceptions to the "acceptable" thresholds listed above, SLA's typically reflect PENALTIES ($$$).


Latency ... the heart of a Bottleneck ...

Each of the attributes and perceived guages of performance listed above has it's own intrinsic relationships and dependencies to specific subsystems and components... in turn reflecting a type of "latency" (delay in response). It is these latencies that are investigated and examined for root cause and correlation as the basis for most Performance Analysis activities.

\*\* STAY TUNED \*\*..  Look for my up-coming blog entry on "The many Flavors of system Latency..".

Future blog entries will expand upon this baseline definition of performance.. so keep your eye's peeled.. and look at the world around you.. from as many vantage points as possible... Perspective is key.. hand in hand with understanding the world around us... Don't be afraid to ask why.. and dig deeper.. there's typically a reason for everything if you look at it with an open mind.. understanding the fundamentals first !

Todd ;) :)

(Copyright 2007, Todd A. Jobson)

Add to Technorati Favorites


This blog does not reflect the viewpoint or opinions of Oracle. All comments are personal reflections and responsibility of Todd A. Jobson, Sr. and are implicitly copyrighted from the posted year to current year, to that effect.


« August 2016