(1) AWR report shows bogus wait events and times on SPARC T5 servers
Here is a sample from one of the Oracle 11g R2 databases running on a SPARC T5 server with Solaris 11.1 SRU 7.5
Top 5 Timed Foreground Events
Avg wait (ms)
% DB time
latch: cache buffers chains
library cache: mutex X
buffer busy waits
latch: In memory undo latch
Reason: Unknown. There is a pending bug 17214885 - Implausible top foreground wait times reported in AWR report.
Tentative workaround: Disable power management as shown below.
# poweradm set administrative-authority=none
# svcadm disable power
# svcadm enable power
Verify the setting by running poweradm list.
Also disable NUMA I/O object binding by setting the following parameter in /etc/system (requires a system reboot).
Oracle Solaris 11 added support for NUMA I/O architecture. Here is a brief explanation of NUMA I/O from Solaris 11 : What's New web page.
Non-Uniform Memory Access (NUMA) I/O : Many modern systems are based on a NUMA architecture, where each CPU or set of CPUs is associated with its own physical memory and I/O devices. For best I/O performance, the processing associated with a device should be performed close to that device, and the memory used by that device for DMA (Direct Memory Access) and PIO (Programmed I/O) should be allocated close to that device as well. Oracle Solaris 11 adds support for this architecture by placing operating system resources (kernel threads, interrupts, and memory) on physical resources according to criteria such as the physical topology of the machine, specific high-level affinity requirements of I/O frameworks, actual load on the machine, and currently defined resource control and power management policies.
Do not forget to rollback these changes after applying the fix for the database bug 17214885, when available.
(2) Redo logs on F40 PCIe cards(non-volatile flash storage)
Per the F40 PCIe card user's guide, the Sun Flash Accelerator F40 PCIe Card is designed to provide best performance for data transfers that are multiples of 8k size, and using addresses that are 8k aligned. To achieve optimal performance, the size of the read/write data should be an integer multiple of this block size and the data transferred should be block aligned. I/O operations that are not block aligned and that do not use sizes that are a multiple of the block size may suffer performance degration, especially for write operations.
Oracle redo log files default to a block size that is equal to the physical sector size of the disk, typically 512 bytes. And most of the time, database writes to the redo log in a normal functioning environment. Oracle database supports a maximum block size of 4K for redo logs. Hence to achieve optimal performance for redo write operations on F40 PCIe cards, tune the environment as shown below.
[Solaris only][F20] To enable maximum throughput from the MPT driver, append the following line to /kernel/drv/mpt.conf and reboot the system.
This tip is applicable to all kinds of flash storage that Oracle sells or sold including F20/F40 PCIe cards and F5100 storage array. sd-config-list in sd.conf may need some adjustment to reflect the correct vendor id and product id.
A NULL value for SCHEMA recompiles all invalid objects in the database.
2. SGA breakdown in Solaris Locality Groups (lgroup)
To find the breakdown, execute pmap -L | grep shm. Then separate the lines that are related to each locality group and sum up the value in 2nd column to arrive at a number that shows the total SGA memory allocated in that locality group.
(I'm pretty sure there will be a much easier way that I am not currently aware of.)
3. Default values for shared pool, java pool, large pool, ..
If the *pool parameters were not set explicitly, executing the following query is one way to find out what are they currently set to.
SQL> select * from v$sgainfo;
NAME BYTES RES
-------------------------------- ---------- ---
Fixed SGA Size 2171296 No
Redo Buffers 373620736 No
Buffer Cache Size 8.2410E+10 Yes
Shared Pool Size 1.7180E+10 Yes
Large Pool Size 536870912 Yes
Java Pool Size 1879048192 Yes
Streams Pool Size 268435456 Yes
Shared IO Pool Size 0 Yes
Granule Size 268435456 No
Maximum SGA Size 1.0265E+11 No
Startup overhead in Shared Pool 2717729536 No
Free SGA Memory Available 0
12 rows selected.
4. Fix to PLS-00201: identifier 'GV$SESSION' must be declared error
Grant select privilege on gv_$SESSION to the owner of the database object that failed to compile.
SQL> alter package OWF_MGR.FND_SVC_COMPONENT compile body;
Warning: Package Body altered with compilation errors.
SQL> show errors
Errors for PACKAGE BODY OWF_MGR.FND_SVC_COMPONENT:
390/22 PL/SQL: Item ignored
390/22 PLS-00201: identifier 'GV$SESSION' must be declared
SQL> grant select on gv_$SESSION to OWF_MGR;
SQL> alter package OWF_MGR.FND_SVC_COMPONENT compile body;
Package body altered.
5. Solaris Critical Thread (CT) optimization for Oracle logwriter (lgrw)
Critical Thread is a new scheduler optimization available in Oracle Solaris releases Solaris 10 Update 10 and later versions. Latency sensitive single threaded components of software such as Oracle database's logwriter benefit from CT optimization.
On a high level, LWPs marked as critical will be granted more exclusive access to the hardware. For example, on SPARC T4 and T5 systems, such a thread will be assigned exclusive access to a core as much as possible. CT optimization won't delay scheduling of any runnable thread in the system.
Critical Thread optimization is enabled by default. However the users of the system have to hint the OS by marking a thread or two "critical" explicitly as shown below.
From database point of view, logwriter (lgwr) is one such process that can benefit from CT optimization on Solaris platform. Oracle DBA's can either make the lgwr process 'critical' once the database is up and running, or can simply patch the 18.104.22.168 database software by installing RDBMS patch 12951619 to let the database take care of it automatically. I believe Oracle 12c does it by default. Future releases of 11g software may make lgwr critical out of the box.
Those who install the database patch 12951619 need to carefully follow the post installation steps documented in the patch README to avoid running into unwanted surprises.
6. ORA-14519 error while importing a table from a Data Pump export dump
All partitions in table XYZ are using 32K blocks whereas the implicit default partition is pointing to a 8K block tablespace. Workaround is to use the REMAP_TABLESPACE option in Data Pump impdp command line to remap the implicit default tablespace of the partitioned table to the tablespace where the rest of partitions are residing.
7. Index building task in Data Pump import process
When Data Pump import process is running, by default, index building is performed with just one thread, which becomes a bottleneck and causes the data import process to take a long time especially if many large tables with millions of rows are being imported into the target database. One way to speed up the import process execution is by skipping index building as part of data import task with the help of EXCLUDE=INDEX impdp command line option. Extract the index definitions for all the skipped indexes from the Data Pump dump file as shown below.
Assembler (as) is not installed by default on Solaris 11.
Find and install
# pkg search assembler
INDEX ACTION VALUE PACKAGE
pkg.fmri set solaris/developer/assembler pkg:/email@example.com
# pkg install pkg:/developer/assembler
Assembler binary used to be under /usr/ccs/bin directory on Solaris 10 and prior versions.
There is no /usr/ccs/bin on Solaris 11. Contents were moved to /usr/bin
2. Non-interactive retrieval of the entire list of disks that format reports
If the format utility cannot show the entire list of disks in a single screen on stdout, it shows some and prompts user to - hit space for more or s to select - to move to the next screen to show few more disks. Run the following command(s) to retrieve the entire list of disks in a single shot.
format < /dev/null
echo "\n" | format
3. Finding system wide file descriptors/handles in use
Run the following kstat command as any user (privileged or non-privileged).
kstat -n file_cache -s buf_inuse
Going through /proc (process filesystem) is less efficient and may lead to inaccurate results due to the inclusion of duplicate file handles.
4. ssh connection to a Solaris 11 host fails with error Couldn't agree a client-to-server cipher (available: aes128-ctr,aes192-ctr,aes256-ctr,arcfour128,arcfour256,arcfour)
Solution: add 3des-cbc to the list of accepted ciphers to sshd configuration file.
Oracle Internet Directory (OID) is an LDAP v3 Directory Server that has multi-threaded, multi-process, multi-instance process architecture with Oracle database as the directory store.
BENCHMARK WORKLOAD DESCRIPTION
Five test scenarios were executed in this benchmark - each test scenario performing a different type of LDAP operation. The key metrics are throughput -- the number of operations completed per second, and latency -- the time it took in milliseconds to complete an operation.
TEST SCENARIOS & RESULTS
1. LDAP Search operation : search for and retrieve specific entries from the directory
In this test scenario, each LDAP search operation matches a single unique entry. Each Search operation results in the lookup of an entry in such a way that no client looks up the same entry twice and no two clients lookup the same entry, and all entries are looked-up randomly.
2. LDAP Add operation : add entries, their object classes, attributes and values to the directory
In this test scenario, 16 concurrent LDAP clients added 500,000 entries of object class InetOrgPerson with 21 attributes to the directory.
3. LDAP Compare operation : compare a given attribute value to the attribute value in a directory entry
In this test scenario, userpassword attribute was compared. That is, each LDAP Compare operation matches user password of a user.
4. LDAP Modify operation : add, delete or replace attributes for entries
In this test scenario, 50 concurrent LDAP clients updated a unique entry each time and a total of 50 million entries were updated. Attribute that is being modified was not indexed
5. LDAP Authentication operation : authenticates the credentials of a user
In this test scenario, 1000 concurrent LDAP clients authenticated 50 million users.
BONUS: LDAP Mixed operations Test
In this test scenario, 1000 LDAP clients were used to perform LDAP Search, Bind and Modify operations concurrently. Operation breakdown (load distribution): Search: 65%. Bind: 30%. Modify: 5%
And finally, the:
1 x Oracle SPARC T5-2 Server
» 2 x 3.6 GHz SPARC T5 sockets each with 16 Cores (Total Cores: 32) and 8 MB L3 cache
» 512 GB physical memory
» 2 x 10 GbE cards
» 1 x Sun Storage F5100 Flash Array with 80 flash modules
» Oracle Solaris 11.1 operating system
Major credit goes to our colleague, Ramaprakash Sathyanarayan
Hardly six months after announcing Siebel 22.214.171.124 benchmark results on Oracle SPARC T4 servers, we have a brand new set of Siebel 126.96.36.199 benchmark results on Oracle SPARC T5 servers. There are no updates to the Siebel benchmark kit in the last couple years - so, we continued to use the Siebel 188.8.131.52 benchmark workload to measure the performance of Siebel Financial Services Call Center and Order Management business transactions on the recently announced SPARC T5 servers.
The latest Siebel 184.108.40.206 benchmark was executed on a mix of SPARC T5-2, SPARC T4-2 and SPARC T4-1 servers. The benchmark test simulated the actions of a large corporation with 40,000 concurrent active users. To date, this is the highest user count we achieved in a Siebel benchmark.
User Load Breakdown & Achieved Throughput
Siebel Application Module
Business Trx per Hour
Financial Services Call Center
Average Transaction Response Times for both Financial Services Call Center and Order Management transactions were under one second.
Software & Hardware Specification
Per Server Specification
Solaris 10 1/13 (S10U11)
Oracle 11g R2
Solaris 10 8/11 (S10U10)
iPlanet Web Server
7.0.9 (7 U9)
Solaris 10 8/11 (S10U10)
Oracle Application Test Suite
AMD Opteron 285 SE
Windows 2003 R2 SP2
Load Drivers (Agents)
Oracle Application Test Suite
Intel Xeon X5670
Windows 2003 R2 SP2
Siebel Gateway Server was configured to run on one of the application server nodes
Four Siebel application servers were configured in the Siebel Enterprise to handle 40,000 concurrent users
- Each SPARC T5-2 was configured to run two Siebel application server instances
- Each of the Siebel application server instances on SPARC T5-2 servers were separated using Solaris virtualization technology, Zones
- 40,000 concurrent user sessions were load balanced across all four Siebel application server instances
Siebel database was hosted on a Sun Storage F5100 Flash Array consisting 80 x 24 GB flash modules (FMODs)
- Siebel 220.127.116.11 benchmark workload is not I/O intensive and does not require flash storage for better I/O performance
Fourteen iPlanet Web Server virtual servers were configured with Siebel Web Server Extension (SWSE) plug-in to handle 40,000 concurrent user load
- All fourteen iPlanet Web Server instances forwarded HTTP requests from Siebel clients to all four Siebel application server instances in a round robin fashion
Oracle Application Test Suite (OATS) was stable and held up amazingly well over the entire duration of the test run.
- The test ran for more than five hours including a three hour ramp up state
The benchmark test results were validated and thoroughly audited by the Siebel benchmark and PSR teams
- Nothing new here. All Sun published Siebel benchmarks including the SPARC T4 one were properly audited before releasing those to the outside world
Finally, how does this benchmark stack up against other published benchmarks? Short answer is "very well". Head over to the Oracle Siebel Benchmark White Papers webpage to do the comparison yourself.
[Credit to our hard working colleagues in SAE, Siebel PSR, benchmark and Oracle Platform Integration (OPI) teams. Special thanks to Sumti Jairath and Venkat Krishnaswamy for the last minute fire drill]
The focus of the paper is on databases and zones. On SuperCluster, customers have the choice of running their databases in logical domains that are dedicated to running Oracle Database 11g R2. With exclusive access to Exadata Storage Servers, those domains are aptly called "Database" domains. If the requirement mandates, it is possible to create and use all logical domains as "database domains" or "application domains" or a mix of those. Since the focus is on databases, the paper talks only about the database domains and how zones can be created, configured and used within each database domain for fine grained control over multiple databases consolidated in a SuperCluster environment.
When multiple databases are being consolidated (including RAC databases) in database logical domains, zones are one of the options that fulfill requirements such as the fault, operation, network, security and resource isolation, multiple RAC instances in a single logical domain, separate identity and independent manageability for database instances.
The best practices cover the following topics. Some of those are applicable to standalone, non-engineered environments as well.
CPU, memory and disk space allocation
Zone Root on Sun ZFS Storage Appliance
Use of DISM
Use of ZFS filesystem
SuperCluster specific zone deployment tool, ssc_exavm
Exadata Storage Grid (Disk Group) Configuration
Disk Group Isolation
Shared Storage approach
Dedicated Storage Server approach
Resizing Grid Disks
Oracle RAC Configuration Securing the Databases, and
Example Database Consolidation Scenarios
Consolidation example using Half-Rack SuperCluster
Consolidation example using Full-Rack SuperCluster
A large group of experts reviewed the material and provided quality feedback. Hence they deserve credit for their work and time. Listed below are some of those reviewers (sincere apologies if I missed listing any major contributors).
Kesari Mandyam, Binoy Sukumaran, Gowri Suserla, Allan Packer, Jennifer Glore, Hazel Alabado, Tom Daly, Krishnan Shankar, Gurubalan T, Rich long, Prasad Bagal, Lawrence To, Rene Kundersma, Raymond Dutcher, David Brean, Jeremy Ward, Suzi McDougall, Ken Kutzer, Larry Mctintosh, Roger Bitar, Mikel Manitius
Just like the Siebel 8.1.x/SPARC T4 benchmark post, this one too was overdue for at least four months. In any case, I hope the Oracle BI customers already knew about the OBIEE 11g/SPARC T4 benchmark effort. In here I will try to provide few additional / interesting details that aren't covered in the following Oracle PR that was posted on oracle.com on 09/30/2012.
The entire BI middleware stack including the WebLogic 11g Server, OBI Server, OBI Presentation Server and Java Host was installed and configured on a single SPARC T4-4 server consisting four 8-Core 3.0 GHz SPARC T4 processors (total #cores: 32) and 128 GB physical memory. Oracle Solaris 10 8/11 is the operating system.
BI users were authenticated against Oracle Internet Directory (OID) in this benchmark - hence OID software which was part of Oracle Identity Management 18.104.22.168.0 was also installed and configured on the system under test (SUT). Oracle BI Server's Query Cache was turned on, and as a result, most of the query results were cached in OBIS layer, that resulted in minimal database activity making it ideal to have the Oracle 11g R2 database server with the OBIEE database running on the same box as well.
Oracle BI database was hosted on a Sun ZFS Storage 7120 Appliance. The BI Web Catalog was under a ZFS/zpool on a couple of SSDs.
In this benchmark, 25000 concurrent users assumed five different business user roles -- Marketing Executive, Sales Representative, Sales Manager, Sales Vice-president, and Service Manager. The load was distributed equally among those five business user roles. Each of those different BI users accessed five different pre-built dashboards with each dashboard having an average of five reports - a mix of charts, tables and pivot tables - and returning 50-500 rows of aggregated data. The benchmark test scenario included drilling down into multiple levels from a table or chart within a dashboard. There is a 60 second think time between requests, per user.
BI Setup & Test Results
OBIEE 11g 22.214.171.124.0 was deployed on SUT in a vertical scale-out fashion. Two Oracle BI Presentation Server processes, one Oracle BI Server process, one Java Host process and two instances of WebLogic Managed Servers handled 25,000 concurrent user sessions smoothly. This configuration resulted in a sub-second overall average transaction response time (average of averages over a duration of 120 minutes or 2 hours). On average, 450 business transactions were executed per second, which triggered 750 SQL executions per second.
It took only 52% of CPU on average (~5% system CPU and rest in user land) to do all this work to achieve the throughput outlined above. Since 25,000 unique test/BI users hammered different dashboards consistently, not so surprisingly bulk of the CPU was spent in Oracle BI Presentation Server layer, which took a whopping 29%. BI Server consumed about 10-11% and the rest was shared by Java Host, OID, WebLogic Managed Server instances and the Oracle database.
So, what is the key take away from this whole exercise?
SPARC T4 rocks Oracle BI world. OBIEE 11g/SPARC T4 is an ideal combination that may work well for majority of OBIEE deployments on Solaris platform. Or in marketing jargon - The excellent vertical and horizontal scalability of the SPARC T4 server gives customer the option to scale up as well as scale out growth, to support large BI EE installations, with minimal hardware investment.
Evaluate and decide for yourself.
[Credit to our colleagues in Oracle FMW PSR, ISVe teams and SCA lab support engineers]
Siebel is a multi-threaded native application that performs well on Oracle's T-series SPARC hardware. We have several versions of Siebel benchmarks published on previous generation T-series servers ranging from Sun Fire T2000 to Oracle SPARC T3-4. So, it is natural to see that tradition extended to the current genration SPARC T4 as well.
29,000 user Siebel 126.96.36.199 benchmark on a mix of SPARC T4-1 and T4-2 servers was announced during the Oracle OpenWorld 2012 event. In this benchmark, Siebel application server instances ran on three SPARC T4-2/Solaris 10 8/11 systems where as the Oracle database server 11gR2 was configured on a single SPARC T4-1/Solaris 11 11/11 system. Several iPlanet web server 7 U9 instances with the Siebel Web Plug-in (SWE) installed ran on one SPARC T4-1/Solaris 10 8/11 system. Siebel database was hosted on a single Sun Storage F5100 flash array consisting 80 flash modules (FMODs) each with capacity 24 GB.
Siebel Call Center and Order Management System are the modules that were tested in the benchmark. The benchmark workload had 70% of virtual users running Siebel Call Center transactions and the remaining 30% vusers running Siebel Order Management System transactions. This benchmark on T4 exhibited sub-second response times on average for both Siebel Call Center and Order Management System modules.
Load balancing at various layers including web and test client systems ensured near uniform load across all web and application server instances. All three Siebel application server systems consumed ~78% CPU on average. The database and web server systems consumed ~53% and ~18% CPU respectively.
All these details are supposed to be available in a standard Oracle|Siebel benchmark template - but for some reason, I couldn't find it on Oracle's Siebel Benchmark White Papers web page yet. Meanwhile check out the following PR that was posted on oracle.com on 09/28/2012.