Monday Jun 30, 2014

Programming in C: Few Tidbits

.. with little commentary aside. Target audience: new programmers. These tips are equally applicable in C and C++ programming environments.

1. Duplicating a file pointer

Steps: find the integer file descriptor associated with the file stream using fileno() call, make a copy of the file descriptor using dup() call, and finally associate the file stream with the duplicated file descriptor by calling fdopen().

FILE *fptr = fopen("file", "mode");

FILE *fptrcopy = fdopen( dup( fileno(fptr) ), "mode");

2. Capturing the exit code of a command that was invoked using popen()

Using pipes is one way of executing commands programmatically that are otherwise invoked from a shell. While pipes are useful in performing tasks other than executing shell commands, this tip is mainly about the exit code of a command (to figure out whether it succeeded or failed) that was executed using popen() API.

To capture the exit code, simply use the value returned by pclose(). This function call returns the termination status of the command that was executed as a child process. However the termination status of the child process is in the top 16 bits of the return value, so dividing the pclose() return value by 256 gives the actual exit code of the command that was executed.

FILE *ptr;
int rc;

if ((ptr = popen("ls", "r")) != NULL) {
	rc = pclose(ptr)/256;
	printf("\nls: exit code = %d", rc);

if ((ptr = popen("ls -W", "r")) != NULL) {
	rc = pclose(ptr)/256;
	printf("\nls -W: exit code = %d", rc);

% ./<executable>

ls: exit code = 0
ls: illegal option -- W
ls -W: exit code = 2

3. Converting an integer to a string

Standard C library has implementation for converting a string to an integer (atoi()), but not for converting an integer to a string. One way to achieve the desired result is by using sprintf() function call, which writes formatted data to a string.

int weight = 30;
char *wtstr = malloc(sizeof(char) * 3);

sprintf(wtstr, "%d", weight);

sprintf() can also be used to convert data in other data types such as float, double to string. Also see: man page for snprintf().

4. Finding the length of a statically allocated array

When size was not specified explicitly, simply divide the total size of the array by the size of the first array element.

static const char *greeting[] = { "Hi", "Hello", "Hola", "Bonjour", \
                                    "Namaste", "Ciao", "Ni Hao" };
int numgreetings = sizeof(greeting)/sizeof(greeting[0]);

After execution, numgreetings variable holds a value of 7. Note that sizeof(greeting[0]) is actually the size of a pointer to a character array.

  • sizeof is not a function, but an operator -- hence it is not necessary or required to use parentheses when using it.
  • Though not so useful, this is applicable even when the size was explicitly specified.

Saturday May 10, 2014

Solaris 11.2 Highlights [Part 2] in 4 Minutes or Less

Part 1: Solaris 11.2 Highlights in 6 Minutes or Less

Highlights contd.,

Package related ..

Minimal Set of System Packages

For the past few years, it is one of the hot topics -- what is the bare minimum [set of packages] needed to run applications. There were a number of blog posts and few technical articles around creating minimal Solaris configurations. Finally users/customers who wish to have their OS installed with minimal set of required system packages for running most of the applications in general, can just install solaris-minimal-server package and not worry about anything else such as removing unwanted packages.

# pkg install pkg:/group/system/solaris-minimal-server

Oracle Database Pre-requisite Package

Until Solaris 11.1, it is up to the users to check the package dependencies and make sure to have those installed before attempting to install Oracle database software especially using graphic installer. Solaris 11.2 frees up the users from the burden of checking and installing individual [required] packages by providing a brand new package called oracle-rdbms-server-12cR1-preinstall. Users just need to install this package for a smoother database software installation later.

# pkg install pkg:/group/prerequisite/oracle/oracle-rdbms-server-12cR1-preinstall

Mirroring a Package Repository

11.2 provides the ability to create local IPS package repositories and keeps them in synch with the IPS package repositories hosted publicly by Oracle Corporation. The key in achieving this is the SMF service svc:/application/pkg/mirror. The following webpage has the essential steps listed on a high-level.

How to Automatically Copy a Repository From the Internet

Another enhancement is the cloning of a package repository using --clone option of pkgrecv command.

Observability related ..

Network traffic diagnostics:

A brand new command, ipstat(1M), reports IP traffic statistics.

# ipstat -?
Usage:	ipstat [-cmnrt] [-a address[,address...]] [-A address[,address...]]
[-d d|u] [-i interface[,interface...]] [-l nlines] [-p protocol[,protocol...]]
[-s key | -S key] [-u R|K|M|G|T|P] [-x opt[=val][,opt[=val]...]]

# ipstat -uM 5

SOURCE                     DEST                       PROTO    INT        BYTES TCP      net8       76.3M TCP      net8        0.6M UDP      net8        0.0M                UDP      net20       0.0M

Total: bytes in: 76.3M bytes out:  0.6M

Another new command, tcpstat(1M), reports TCP and UDP traffic statistics.

# tcpstat -?
Usage:	tcpstat [-cmnrt] [-a address[,...]] [-A address[,...]] [-d d|u] [-i pid[,...]] 
[-l nlines] [-p port[,...]] [-P port[,...]] [-s key | -S key] [-x opt[=val][,...]] 
[-z zonename[,...]] [interval [count]]

# tcpstat 5

ZONE         PID PROTO  SADDR             SPORT DADDR             DPORT   BYTES
global      1267 TCP  42972 etc2m-appadm01.u     22   84.3M
global      1267 TCP    etc2m-appadm01.u     22  42972   48.0K
global      1089 UDP      161    33436  137.0 
global      1089 UDP    33436      161   44.0 

Total: bytes in: 84.3M bytes out: 48.4K

# tcpstat -i 43982 5		<-- TCP stats for a given pid

ZONE         PID PROTO  SADDR             SPORT DADDR             DPORT   BYTES
global     43982 TCP    etc2m-appadm01.u  43524     22   73.7M
global     43982 TCP     22 etc2m-appadm01.u  43524   41.9K

Total: bytes in: 42.1K bytes out: 73.7M

Up until 11.1, it is not so straight-forward to figure out what process created a network endpoint -- one has to rely on a combination of commands such as netstat, pfiles or lsof and proc filesystem (/proc) to extract that information. Solaris 11.2 attempts to make it easy by enhancing the existing tool netstat(1M). Enhanced netstat(1M) shows what user, pid created and control a network endpoint. -u is the magic flag.

#  netstat -aun			<-- notice the -u flag in netstat command; and User, Pid, Command columns in the output

   Local Address        Remote Address      User    Pid      Command       State
-------------------- -------------------- -------- ------ -------------- ----------
      *.*                                 root        162 in.mpathd      Unbound
      *.*                                 netadm      765 nwamd          Unbound
      *.55388                             root        805 picld          Idle

   Local Address        Remote Address      User     Pid     Command     Swind  Send-Q  Rwind  Recv-Q    State
-------------------- -------------------- -------- ------ ------------- ------- ------ ------- ------ ----------- root       1267 sshd           128872      0  128872      0 ESTABLISHED    root          0       2094176      0 1177974      0 ESTABLISHED            *.*            root       2943 nmz                 0      0 1048576      0 LISTEN             *.*            pkg5srv   16012 httpd.worker        0      0 1048576      0 LISTEN

[x86 only] Memory Access Locality Characterization and Analysis

Solaris 11.2 introduced another brand new tool, numatop(1M), that helps in characterizing the NUMA behavior of processes and threads on systems with Intel Westmere, Sandy Bridge and Ivy Bridge processors. If not installed by default, install the numatop package as shown below.

# pkg install pkg:/diagnostic/numatop

Performance related ..

This is a grey area - so, just be informed that there are some ZFS and Oracle database related performance enhancements.

Starting with 11.2, ZFS synchronous write transactions are committed in parallel, which should help improve the I/O throughput.

Database startup time has been greatly improved in Solaris 11 releases -- it's been further improved in 11.2. Customers with databases that use hundreds of Gigabytes or Terabyte(s) of memory will notice the improvement to the database startup times. Other changes to asynchronous I/O, inter-process communication using event ports etc., help improve the performance of the recent releases of Oracle database such as 12c.

Miscellaneous ..

Java 8

Java 7 is still the default in Solaris 11.2 release, but Java 8 can be installed from the IPS package repository.


# pkg install pkg:/developer/java/jdk-8		<-- Java Development Kit
# pkg install pkg:/runtime/java/jre-8		<-- Java Runtime

Bootable USB Media

Solaris 11.2 introduces the support for booting SPARC systems from USB media. Use Solaris Distribution Constructor (requires distribution-constructor package) to create the USB bootable media, or copy a bootable/installation image to the USB media using usbcopy(1M) and dd(1M) commands.

Oracle Hardware Management Pack

Oracle Hardware Management Pack is a set of tools that are integrated into the Solaris OS distribution, that show the existing hardware configuration, help configure hardware RAID volumes, update server firmware, configure ILOM service processor, enable monitoring the hardware using existing tools etc., Look for pkg:/system/management/hmp/hmp-* packages.

Few other interesting packages:

Parallel implementation of bzip2 : compress/pbzip2
NVM Express (nvme) utility : system/storage/nvme-utilities
Utility to administer cluster of servers : terminal/cssh

Tuesday Apr 29, 2014

Solaris 11.2 Highlights [Part 1] in 6 Minutes or Less

This is not the complete list, of course. Just a few hand-picked ones.

First things first, Solaris 11.2 beta is out.

URLs: Download | What's New in Solaris 11.2 | Information Library (documentation)


Zones related ..

Kernel Zones

Kernel Zones bring the ability to run a non-global/local zone at a different kernel version from the global zone and can be patched or updated independently without the need to reboot the global zone. In other words, kernel zones are independent and isolated environments with a full kernel and user environment.

Creating a Kernel Zone:

  1. If not available, install the kernel zone brand
    # pkg install brand/brand-solaris-kz
  2. Create and install a kernel zone using the existing zonecfg and zoneadm commands. The only difference compared to creating a non-kernel zone (the zones we have been creating for the past 10 years) is the template to be used -- by default, SYSdefault template is used. To create a kernel zone, use SYSsolaris-kz template instead.

    # zonecfg -z <zone-name> create –t SYSsolaris-kz
    # zoneadm –z <zone-name> install
    # .. continue with the rest of the steps to complete zone configuration ..

Kernel Zones can be used in combination with logical domains (Oracle VM for SPARC), but cannot be used in combination with other virtualization solutions such as Oracle VM VirtualBox that does not support nested virtualization.

Live Zone Re-configuration

This release (11.2) added support for the dynamic re-configuration of local zones. Now the following configuration changes do not require a zone reboot.

  • Resource controls and pools
  • Network configuration
  • Adding or removing file systems
  • Adding or removing virtual and physical devices

Read-Only Global Zones

Recent releases of Solaris have support for Immutable Non-Global Zones already. Solaris 11.2 extends the immutable zone support to Global Zones. Immutable zones will have a read-only zone root.

Make a Global Zone Read-Only/Immutable by:

# zonecfg -z global set file-mac-profile=fixed-configuration

Installing Packages across multiple Non-Global Zones from the Global Zone

  • -r option of pkg can be used to install/update/uninstall software packages into/in/from all non-global zones from the global zone.
  • Use -Z option along with -r to exclude a zone in applying the package operation. Similarly use -z along with -r to apply the intended package operation only in a specific zone

Multiple Boot Environments for Solaris 10 Zones

Multiple BE support has been extended to Solaris 10 Zones in this release. This feature is useful when performing operations such as patching within an Solaris 10 environment running on a Solaris 11 system.

CMT Aware Zones and Resource Pool Configuration

It is now possible to allocate CMT based resources -- vCPUs, Cores and Sockets, using the existing zonecfg and poolcfg commands. This is useful from performance and/or licensing point of view as it provides flexibility and control for managing licensing boundaries or dedicating hardware resources solely to a zone.

Cloud related ..

Centralized Cloud Management with OpenStack

Solaris 11.2 is the first release to incorporate a complete OpenStack distribution. OpenStack allows managing and sharing compute, network and storage resources in the data center through a centralized web portal. In other words, now administrators can set up an enterprise ready private cloud Infrastructure-as-a-Service (IaaS) environment with ease.

Check this quick How-To article out at Oracle Technology Network -- Getting Started with OpenStack on Oracle Solaris 11.2

Cloning and Disaster Recovery with Unified Archives

Unified Archives is a new native archive type that enables quick cloning for rapid application deployment in the cloud, fast and reliable disaster recovery. Both bare metal and virtual environments are supported. Check the archiveadm(1M) man page for details.

Create a clone archive of a system
# archiveadm create ./clone.uar

Create bootable media
# archiveadm create-media ./archive.uar				/* USB image */
# archiveadm create-media -f iso <other options> ./bootarch.uar	/* ISO image */

Create a full system recovery archive
# archiveadm create --recovery ./recovery.uar

Extract information from a Unified Archive
# archiveadm info somearchive.uar

To be continued .. Stay tuned.

Monday Mar 31, 2014

[Solaris] ZFS Pool History, Writing to System Log, Persistent TCP/IP Tuning, ..

.. with plenty of examples and little comments aside.

[1] Check existing DNS client configuration

Solaris 11 and later:

% svccfg -s network/dns/client listprop config
config                      application        
config/value_authorization astring
config/options             astring     "ndots:2 timeout:3 retrans:3 retry:1"
config/search              astring     "" "" "" "" ""
config/nameserver          net_address

Solaris 10 and prior:

Check the contents of /etc/resolv.conf

% cat /etc/resolv.conf
options ndots:2 timeout:3 retrans:3 retry:1

Note that /etc/resolv.conf file exists on Solaris 11.x releases too as of today.

[2] Logical domains: finding out the hostname of control domain

Use virtinfo(1M) command.

root@ppst58-cn1-app:~# virtinfo -a
Domain role: LDoms guest I/O service root
Domain name: n1d2
Domain UUID: 02ea1fbe-80f9-e0cf-ecd1-934cf9bbeffa
Control domain: ppst58-01
Chassis serial#: AK00083297

The above output shows that n1d2 domain is a guest domain, which is also an I/O domain, the service domain and a root I/O domain. Control domain is running on host ppst58-01.

Output from control domain:

root@ppst58-01:~# ldm list
primary          active     -n-cv-  UART    64    130304M  0.1%  0.1%  243d 2h 
n1d1             active     -n----  5001    448   916992M  0.2%  0.2%  3d 15h 26m
n1d2             active     -n--v-  5002    512   1T       0.0%  0.0%  3d 15h 29m

root@ppst58-01:~# virtinfo -a
Domain role: LDoms control I/O service root
Domain name: primary
Domain UUID: 19337210-285a-6ea4-df8f-9dc65714e3ea
Control domain: ppst58-01
Chassis serial#: AK00083297

[3] Administering NFS configuration

Solaris 11 and later:

Use sharectl(1M) command. Solaris 11.x releases include the sharectl administrative tool to configure and manage file-sharing protocols such as NFS, SMB, autofs.

Display all property values of NFS:

# sharectl get nfs

# sharectl status
autofs  online client
nfs     disabled

Modifying the nfs v4 grace period from the default 90s to 30s:

# sharectl get -p grace_period nfs
# sharectl set -p grace_period=30 nfs
# sharectl get -p grace_period nfs

Solaris 10 and prior:

Edit /etc/default/nfs file, and restart NFS related service(s).

[4] Examining ZFS Storage Pool command history

Solaris 10 8/07 and later releases log successful zfs and zpool commands that modify the underlying pool state. All those executed commands can be examined by running zpool history command. Because this command shows the actual zfs commands executed as they are, the 'history' feature is really useful in troubleshooting an error scenario that was resulted from executing some zfs command.

# zpool list
rpool      416G   152G  264G  36%  1.00x   ONLINE  -
zs3actact  848G  17.4G  831G   2%  1.00x   ONLINE  -

# zpool history -l zs3actact
History for 'zs3actact':
2014-03-19.22:02:32 zpool create -f zs3actact c0t600144F0AC6B9D2900005328B7570001d0 [user root on etc25-appadm05:global]
2014-03-19.22:03:12 zfs create zs3actact/iscsivol1 [user root on etc25-appadm05:global]
2014-03-19.22:03:33 zfs set recordsize=128k zs3actact/iscsivol1 [user root on etc25-appadm05:global]

Note that this log is enabled by default, and cannot be disabled.

[5] Modifying TCP/IP configuration parameters

Using ndd(1M) is the old way of tuning TCP/IP parameters, and still supported as of today (in Solaris 11.x releases). However using padm(1M) command is the recommended way to modify or retrieve TCP/IP Internet protocols on Solaris 11.x and later releases.

# ipadm show-prop -p max_buf tcp
tcp   max_buf               rw   1048576      --           1048576      128000-1073741824

# ipadm set-prop -p max_buf=2097152 tcp

# ipadm show-prop -p max_buf tcp
tcp   max_buf               rw   2097152      2097152      1048576      128000-1073741824

ndd style (still valid):

# ndd -get /dev/tcp tcp_max_buf

# ndd -set /dev/tcp tcp_max_buf 2097152

# ndd -get /dev/tcp tcp_max_buf

One of the advantages of using ipadm over ndd is that the configured/tuned non-default values are persistent across reboots. In case of ndd, we have to re-apply those values either manually or by creating a Run Control script (/etc/rc*.d/S*) to make sure that the intended values are set automatically during a reboot of the system.

[6] Writing to system log from a shell script

Use logger(1) command as shown in the following example.


# logger -p local0.warning Big Brother is watching you

# dmesg | tail -1
Mar 30 18:42:14 etc27zadm01 root: [ID 702911 local0.warning] Big Brother is watching you

Check syslog.conf(4) man page for the list of available system facilities and the severity of the condition being logged (levels).


[*] Forceful NFS unmount on Linux

Try the lazy unmount option (-l) on systems running Linux kernel 2.4.11 or later to forcefully unmount a filesystem that keeps throwing Device or resource busy and/or device is busy error(s).


# umount -f /bkp
umount2: Device or resource busy
umount: /bkp: device is busy
umount2: Device or resource busy
umount: /bkp: device is busy

# umount -l /bkp

Wednesday Mar 26, 2014

Software Availability : Solaris Studio 12.4 Beta & ORAchk

First off, these are two unrelated softwares.

Solaris Studio 12.4 Beta

Nearly two-and-a-half years after the release of Solaris Studio 12.3, Oracle is gearing up for the next major release 12.4. In addition to the compiler and library optimizations to support the latest and greatest SPARC & Intel x64 hardware such as SPARC T5, M5, M6, Fujitsu's M10, and Intel's Ivy Bridge and Haswell line of servers, support for C++ 2011 language standard is one of the highlights of this forthcoming release. The complete list of features and enhancements in release 12.4 are documented in the What's New page.

Those who feel compelled to give the updated/enhanced compilers and tools a try, can get started right away by downloading the beta bits from the following location. This software is available for Solaris 10 & 11 running on SPARC, x86 hardware; and Linux 5 & 6 runnin g on x86/x64 hardware. Anyone can download this software for free.

     Oracle Solaris Studio 12.4 Beta Download

Don't forget to check the Release Notes out for the installation instructions, known issues, limitations and workarounds, features that were removed in this release and so on.

Here's a pointer to the documentation (preview): Oracle Solaris Studio 12.4 Information Library

Finally, should you run into any issue(s) or if you have questions about anything related, feel free to use the Solaris Studio Community Forum.

ORAchk 2.2.4 (formerly known as EXAchk)

ORAchk, the Oracle Configuration Audit Tool, enhances EXAchk tool's functionality, and replaces the existing & popular RACcheck tool. In addition to the top issues reported by users/customers, ORAchk proactively scans for known problems within Oracle Database, Sun systems (especially engineered systems) and Oracle E-Business Suite Financials.

While checking, ORAchk covers a wide range of areas such as OS kernel settings, database installations (single instance and RAC), performance, backup and recovery, storage setting, and so on.

ORAchk generated reports (mostly high level) show the system health risks with the ability to drill down into specific problems and offers recommendations specific to the environment and product configuration. Those who do not like sending this data back to Oracle should be happy to know that there is no phone home feature in this release.

Note that ORAchk is available only for the Oracle Premier Support Customers - meaning only those customers with appropriate support contracts can use this tool. So, if you are a Oracle customer with the ability to access the Oracle Support website, check the following pages out for additional information.

     ORAchk - Oracle Configuration Audit Tool
     ORAchk user's guide

Feel free to use the community forum to ask any related questions.

Friday Feb 28, 2014

[Solaris] Changing hostname, Parallel Compression, pNFS, Upgrading SRUs and Clearing Faults

[1] Solaris 11+ : changing hostname

Starting with Solaris 11, a system's identify (nodename) is configured through the config/nodename service property of the svc:/system/identity:node SMF service. Solaris 10 and prior versions have this information in /etc/nodename configuration file.

The following example demonstrates the commands to change the hostname from "ihcm-db-01" to "ehcm-db-01".

# hostname

# svccfg -s system/identity:node listprop config
config                       application        
config/enable_mapping       boolean     true
config/ignore_dhcp_hostname boolean     false
config/nodename             astring     ihcm-db-01
config/loopback             astring     ihcm-db-01

# svccfg -s system/identity:node setprop config/nodename="ehcm-db-01"

# svccfg -s system/identity:node refresh  -OR- 
	# svcadm refresh svc:/system/identity:node
# svcadm restart system/identity:node

# svccfg -s system/identity:node listprop config
config                       application        
config/enable_mapping       boolean     true
config/ignore_dhcp_hostname boolean     false
config/nodename             astring     ehcm-db-01
config/loopback             astring     ehcm-db-01

# hostname

[2] Parallel Compression

This topic is not Solaris specific, but certainly helps Solaris users who are frustrated with the single threaded implementation of all officially supported compression tools such as compress, gzip, zip.

pigz (pig-zee) is a parallel implementation of gzip that suits well for the latest multi-processor, multi-core machines. By default, pigz breaks up the input into multiple chunks of size 128 KB, and compress each chunk in parallel with the help of light-weight threads. The number of compress threads is set by default to the number of online processors. The chunk size and the number of threads are configurable.

Compressed files can be restored to their original form using -d option of pigz or gzip tools. As per the man page, decompression is not parallelized out of the box, but may show some improvement compared to the existing old tools.

The following example demonstrates the advantage of using pigz over gzip in compressing and decompressing a large file.


Original file, and the target hardware.

$ ls -lh PT8.53.04.tar 
-rw-r--r--   1 psft     dba         4.8G Feb 28 14:03 PT8.53.04.tar

$ psrinfo -pv
The physical processor has 8 cores and 64 virtual processors (0-63)
  The core has 8 virtual processors (0-7)
  The core has 8 virtual processors (56-63)
    SPARC-T5 (chipid 0, clock 3600 MHz)

gzip compression.

$ time gzip --fast PT8.53.04.tar 

real    3m40.125s
user    3m27.105s
sys     0m13.008s

$ ls -lh PT8.53*
-rw-r--r--   1 psft     dba         3.1G Feb 28 14:03 PT8.53.04.tar.gz

/* the following prstat, vmstat outputs show that gzip is compressing the 
	tar file using a single thread - hence low CPU utilization. */

$ prstat -p 42510

 42510 psft     2616K 2200K cpu16    10    0   0:01:00 1.5% gzip/1

$ prstat -m -p 42510

 42510 psft      95 4.6 0.0 0.0 0.0 0.0 0.0 0.0   0  35  7K   0 gzip/1

$ vmstat 2

 r b w   swap  free  re  mf pi po fr de sr s0 s1 s2 s3   in   sy   cs us sy id
 0 0 0 776242104 917016008 0 7 0 0 0  0  0  0  0 52 52 3286 2606 2178  2  0 98
 1 0 0 776242104 916987888 0 14 0 0 0 0  0  0  0  0  0 3851 3359 2978  2  1 97
 0 0 0 776242104 916962440 0 0 0 0 0  0  0  0  0  0  0 3184 1687 2023  1  0 98
 0 0 0 775971768 916930720 0 0 0 0 0  0  0  0  0 39 37 3392 1819 2210  2  0 98
 0 0 0 775971768 916898016 0 0 0 0 0  0  0  0  0  0  0 3452 1861 2106  2  0 98

pigz compression.

$ time ./pigz PT8.53.04.tar 

real    0m25.111s	<== wall clock time is 25s compared to gzip's 3m 27s
user    17m18.398s
sys     0m37.718s

/* the following prstat, vmstat outputs show that pigz is compressing the 
        tar file using many threads - hence busy system with high CPU utilization. */

$ prstat -p 49734

49734 psft       59M   58M sleep    11    0   0:12:58  38% pigz/66

$ vmstat 2

 kthr      memory            page            disk          faults      cpu
 r b w   swap  free  re  mf pi po fr de sr s0 s1 s2 s3   in   sy   cs us sy id
 0 0 0 778097840 919076008 6 113 0 0 0 0 0  0  0 40 36 39330 45797 74148 61 4 35
 0 0 0 777956280 918841720 0 1 0 0 0  0  0  0  0  0  0 38752 43292 71411 64 4 32
 0 0 0 777490336 918334176 0 3 0 0 0  0  0  0  0 17 15 46553 53350 86840 60 4 35
 1 0 0 777274072 918141936 0 1 0 0 0  0  0  0  0 39 34 16122 20202 28319 88 4 9
 1 0 0 777138800 917917376 0 0 0 0 0  0  0  0  0  3  3 46597 51005 86673 56 5 39

$ ls -lh PT8.53.04.tar.gz 
-rw-r--r--   1 psft     dba         3.0G Feb 28 14:03 PT8.53.04.tar.gz

$ gunzip PT8.53.04.tar.gz 	<== shows that the pigz compressed file is 
                                         compatible with gzip/gunzip

$ ls -lh PT8.53*
-rw-r--r--   1 psft     dba         4.8G Feb 28 14:03 PT8.53.04.tar


$ time ./pigz -d PT8.53.04.tar.gz 

real    0m18.068s
user    0m22.437s
sys     0m12.857s

$ time gzip -d PT8.53.04.tar.gz 

real    0m52.806s <== compare gzip's 52s decompression time with pigz's 18s
user    0m42.068s
sys     0m10.736s

$ ls -lh PT8.53.04.tar 
-rw-r--r--   1 psft     dba         4.8G Feb 28 14:03 PT8.53.04.tar

Of course, there are other tools such as Parallel BZIP2 (PBZIP2), which is a parallel implementation of the bzip2 tool are worth a try too. The idea here is to highlight the fact that there are better tools out there to get the job done in a quick manner compared to the existing/old tools that are bundled with the operating system distribution.

[3] Solaris 11+ : Upgrading SRU

Assuming the package repository is set up already to do the network updates on a Solaris 11+ system, the following commands are helpful in upgrading a SRU.

  • List all available SRUs in the repository.

    # pkg list -af entire
  • Upgrade to the latest and greatest.

    # pkg update

    To find out what changes will be made to the system, try a dry run of the system update.

    # pkg update -nv
  • Upgrade to a specific SRU.

    # pkg update entire@<FMRI>

    Find the Fault Managed Resource Identifier (FMRI) string by running pkg list -af entire command.

Note that it is not so easy to downgrade SRU to a lower version as it may break the system. Should there be a need to downgrade or switch between different SRUs, relying on Boot Environments (BE) might be a good idea. Check Creating and Administering Oracle Solaris 11 Boot Environments document for details.

[4] Parallel NFS (pNFS)

Just a quick note — RFC 5661, Network File System (NFS) Version 4.1 introduced a new feature called "Parallel NFS" or pNFS, which allows NFS clients to access storage devices containing file data directly. When file data for a single NFS v4 server is stored on multiple and/or higher-throughput storage devices, using pNFS can result in significant improvement in file access performance. However Parallel NFS is an optional feature in NFS v4.1. Though there was a prototype made available few years ago when OpenSolaris was still alive, as of today, Solaris has no support for pNFS. Stay tuned for any updates from Oracle Solaris teams.

Here is an interesting write-up from one of our colleagues at Oracle|Sun (dated 2007) -- NFSv4.1's pNFS for Solaris.

(Credit to Rob Schneider and Tom Gould for initiating this topic)

[5] SPARC hardware : Check for and clear faults from ILOM

Couple of ways to check the faults using ILOM command line interface.

By running:

  1. show faulty command from ILOM command prompt, or
  2. fmadm faulty command from within the ILOM faultmgmt shell

Once found, use the clear_fault_action property with the set command to clear the fault for a FRU.

The following example checks for the faulty FRUs from ILOM faultmgmt shell, then clears it out.


-> start /SP/faultmgmt/shell
Are you sure you want to start /SP/faultmgmt/shell (y/n)? y

faultmgmtsp> fmadm faulty

------------------- ------------------------------------ -------------- --------
Time                UUID                                 msgid          Severity
------------------- ------------------------------------ -------------- --------
2014-02-26/16:17:11 18c62051-c81d-c569-a4e6-e418db2f84b4 PCIEX-8000-SQ  Critical
Suspect 1 of 1
   Fault class  :
   Certainty    : 100%
   Affects      : hc:///chassis=0/motherboard=0/cpuboard=1/chip=2/hostbridge=4
   Status       : faulted

      Status            : faulty
      Location          : /SYS/PM1
      Manufacturer      : Oracle Corporation
      Name              : TLA,PM,T5-4,T5-8

Description : A fault has been diagnosed by the Host Operating System.

Response    : The service required LED on the chassis and on the affected
              FRU may be illuminated.


faultmgmtsp> exit

-> set /SYS/PM1 clear_fault_action=True
Are you sure you want to clear /SYS/PM1 (y/n)? y
Set 'clear_fault_action' to 'True'

Note that this procedure clears the fault from the SP but not from the host.

Tuesday Feb 25, 2014

AIX customers: Run for the Hills ..

.. or keep your cool and embrace Solaris.

When Oracle acquired Sun, IBM tried to capitalize the situation just like every other competitor Sun had – doubts raised about Oracle's ability to turn Sun's hardware business around, and Solaris customers were advised to flee SPARC. Fast forward four years .. Oracle appears to have successfully dispelled the doubts with proven long-term commitment to the Solaris/SPARC business with consistent investment and delivery on established roadmaps. Besides, Oracle has been innovating in the server space with engineered systems that are pre-integrated to reduce the cost and complexity of IT infrastructures while increasing productivity and performance.

On the other hand, judging by the recent turn of events at IBM such as selling off critical server technologies, decline in data center business, employee furloughs, layoffs etc., it appears that Big Blue has its own struggles to deal with. In any case, irrespective of what is happening at IBM, AIX customers who are contemplating to migrate to a modern operating platform that is reliable, secure, cloud-ready and offers a rich set of features to virtualize, consolidate, diagnose, debug and most importantly scale and perform, have an attractive alternative — Oracle Solaris. Act before it is too late.

Unfortunately migrating larger deployments from one platform to another is not as easy as migrating desktop users from one operating system to another. So, Oracle put together a bunch of documents to make the AIX to Solaris transition as smooth as possible for the existing AIX customers. Access the AIX-to-Solaris migration pages at:
     Modernizing IBM AIX/Power to Oracle Solaris/SPARC (Oracle Technology Network)

The above pages have pointers to white papers such as IBM AIX to Oracle Solaris Technology Mapping Guide (for system admins, power users), Simplify the Migration of Oracle Database and Oracle Applications from AIX to Oracle Solaris (for DBAs, application specific admins) and IBM AIX Technologies Compared to Oracle Solaris 11 along with hands-on labs, training, blogs and other useful resources. Check those out, and use the contact information available in those pages to speak or chat with relevant Oracle team(s) who can help get started with the migration process. Good luck.

Friday Jan 31, 2014

Solaris Tips : Automounted NFS, ZFS metaslabs, utility to manage F40 cards, powertop, ..

[1] Mounting NFS on Solaris 10 and later

With a relevant entry in /etc/vfstab, the general expectation is that Solaris automatically mounts the NFS shares upon a system reboot. However users may find that NFS shares are not being auto-mounted on some of the systems running the latest update of Solaris 10 or 11. One reason for this behavior could be the use of the Secure By Default network profile, which was introduced in Solaris 10 11/06. When this networking profile is in use, numerous services including the NFS client service are disabled. For the automounting of NFS shares, we will need the NFS client service running.

The fix is to enable NFS client service along with its dependencies.

# svcs -a | grep nfs\/client
disabled       Jan_17   svc:/network/nfs/client:default

# svcadm  enable -r svc:/network/nfs/client

# svcs -a | grep nfs\/client
online         Jan_20   svc:/network/nfs/client:default

On a similar note, if you want all default services to be enabled as they were in previous Solaris releases, run the following command as privileged user. Then use svcadm(1M) to disable unwanted services.

# netservices open

To switch back to the secure by default profile, run:

# netservices limited

[2] Utility to manage Sun Flash Accelerator F40 PCIe card(s) .. ddcli

The Sun Flash Accelerator F40 PCIe Card has two sets of firmware — NAND flash controller firmware, and SAS controller firmware (host PCIe to SAS controller). Both firmware sets are updated as a single F40 firmware package using the ddcli utility. This utility can be used to locate and display information about the cards in the system, format the cards, monitor the health and extract smart logs (to assist Oracle support in debugging and resolution) for a selected F40 card.

If ddcli utility is not available on systems where the F40 PCIe cards are installed, install patch "16005846: F40 (AURA 2) SW1.1 Release fw ( and cli utility update" or later version, if available. This patch can be downloaded from

Note that ddcli utility can be used to service and monitor the health of Sun Flash Accelerator F80 PCIe cards too. Install patch "Patch 17860600: SW1.0 for Sun Flash Acccelerator F80" to get access to the F80 card software package.

[3] Permission denied error when changing a password

An attempt to change the password for a local user 'XYZ' fails with Permission denied error.

# passwd XYZ
New Password: ********
Re-enter new Password: ********
Permission denied

# grep passwd /etc/nsswitch.conf
passwd: files ldap

Users have the flexibility to include and access password information in/from multiple repositories such as files and nis or ldap. Per the man page of passwd(1), when a user has a password stored in one of the name services as well as a local files entry, the passwd command tries to update both. It is possible to have different passwords in the name service and local files entry. Use passwd -r to change a specific password repository.

Hence the fix is to use the -r option in this case to ignore the nsswitch.conf file sequence and update the password information in local /etc files — /etc/passwd and /etc/shadow files.

# passwd -r files XYZ
New Password: ********
Re-enter new Password: ********
passwd: password successfully changed for oracle

[4] Microstate statistics for any process

ptime -m shows the full set of microstate accounting statistics for the lifetime of a given process. prstat -m also reports the microstate process accounting information, but the displayed statistics are accumulated since last display every interval seconds.

# prstat -p 39235

 39235 psft     3585M 3320M sleep    59    0   2:23:11 0.0% java/257

# prstat -mp 39235

 39235 psft     0.0 0.0 0.0 0.0 0.0  87  13 0.0   0   0   1   0 java/257

# ptime -mp 39235

real 428:31:25.902644700
user  2:06:32.283801209
sys     16:37.056999418
trap        2.250539737
tflt        0.000000000
dflt        2.018347218
kflt        0.000000000
lock 96013:52:37.184929717
slp  14349:50:02.286168683
lat      3:11.510473038
stop        0.002468763

In the above example, java process with pid 39235 spent most of its time sleeping waiting to acquire locks in user space (ref: 'lock' field). It also spent a lot of time in just sleeping waiting for some work (ref: 'slp' field). User CPU time is the next major one (ref: 'user' field). The process spent a little bit of time in system space (ref: 'sys' field), waiting for CPU (ref: 'lat' field) and almost negligible amount of time in processing system traps (ref: 'trap' field) and in servicing data page faults (ref: 'dflt' field).

[5] ZFS : metaslab utilization

ZFS divides the space on each device (virtual or physical) into a number of smaller, manageable regions called metaslabs. Each metaslab is associated with a space map that holds information about the free space in that region by keeping tracking of space allocations and deallocations.

The following sample outputs show that a virtual device, u01, made up of two physical disks has 139 metaslabs. The number of segments and free/available space in each metaslab is also shown in those outputs.

# zpool list u01
u01   1.09T   133G  979G  11%  1.00x  ONLINE  -

# zpool status u01
  pool: u01
 state: ONLINE
  scan: none requested

        NAME                       STATE     READ WRITE CKSUM
        u01                        ONLINE       0     0     0
          mirror-0                 ONLINE       0     0     0
            c0t5000CCA01D1DD4A4d0  ONLINE       0     0     0
            c0t5000CCA01D1DCE88d0  ONLINE       0     0     0

errors: No known data errors

# zdb -m u01

        vdev          0   ms_array         27
        metaslabs   139   offset                spacemap          free      
        ---------------   -------------------   ---------------   -------------
        metaslab      0   offset            0   spacemap     30   free    4.65M
        metaslab      1   offset    200000000   spacemap     32   free     698K
        metaslab      2   offset    400000000   spacemap     33   free    1.25M
        metaslab      3   offset    600000000   spacemap     35   free     588K
        metaslab     62   offset   7c00000000   spacemap      0   free       8G
        metaslab     63   offset   7e00000000   spacemap     45   free    8.00G
        metaslab     64   offset   8000000000   spacemap      0   free       8G
        metaslab    136   offset  11000000000   spacemap      0   free       8G
        metaslab    137   offset  11200000000   spacemap      0   free       8G
        metaslab    138   offset  11400000000   spacemap      0   free       8G

# zdb -mm u01   

        vdev          0   ms_array         27
        metaslabs   139   offset                spacemap          free      
        ---------------   -------------------   ---------------   -------------
        metaslab      0   offset            0   spacemap     30   free    4.65M
                          segments       1136   maxsize    103K   freepct    0%
        metaslab      1   offset    200000000   spacemap     32   free     698K
                          segments         64   maxsize    118K   freepct    0%
        metaslab      2   offset    400000000   spacemap     33   free    1.25M
                          segments        113   maxsize    104K   freepct    0%
        metaslab      3   offset    600000000   spacemap     35   free     588K
                          segments        109   maxsize   28.5K   freepct    0%

What is the purpose of this topic? Just to introduce the ZFS debugger, zdb (check the man page zdb(1M)) to the power-users who would like to dig a little deep to find answers to tough questions such as if a ZFS filesystem is fragmented.

Keywords: ZFS zdb metaslab "space map"

[6] Roles can not login directly error on Solaris 11 and later

The root account in Solaris 11 is a role. A role is just like any other user account with the exception that users with roles cannot login directly. Here is an example that shows the failure when attempted to connect directly.

login: root
Password: ********
Roles can not login directly

In this example, connecting as a normal user (who have no roles assigned) and then using su to connect as root user would succeed. This additional step is to prevent malevolent users from getting away with no accountability. Check Bart's blog post SPOTD: The Guide Book to Solaris Role-Based Access Control for some relevant information.

If security is not a primary concern, and if connecting directly as root user is desirable, simply change the root role into a user.

# rolemod -K type=normal root

This change does not affect all the users who are currently in the root role — they retain the root role. Other users who have root access can su to root or log in to the system as the root user. To remove the root role assignment from other local users, set the role to an empty string using usermod command as shown in the following example.

/* assign root role to user 'giri' */
# usermod -R root giri

# roles giri

/* remove the role from user 'giri' */
# usermod -R "" giri

Keywords: RBAC, roles

[7] Large volume sizes (> 2 TB), and maximum size of UFS filesystem

As per the Solaris System Administration Guide, the maximum size of a UFS filesystem is ~16 TB.

To create a UFS file system greater than 2 TB, use EFI disk label. The EFI label provides support for physical disks and virtual disk volumes that are greater than 2 TB in size. Refer to the disk management section in Solaris System Administration Guide to find out the advantages and limitations of EFI.

Note that ZFS labels disks with an EFI label when creating a ZFS storage pool (zpool). And users in general need not be too concerned about the maximum size of a ZFS filesystem as it is several times larger than the maximum size supported by the UFS filesystem.

[8] powertop to observe the CPU power management

Although powertop was ported to Solaris and available as an add-on package from unofficial sources for the past few years, recent releases of Solaris bundled this tool with the core distribution. powertop can be used to monitor the effectiveness of CPU power management features on systems running Solaris. It also displays the clock frequently at which the CPU is operating along with the top events that are causing the CPU to wake up and use more energy.

Be aware that when the CPU power management is enabled with the elastic policy in effect (default on Solaris 11 and later), the CPUs on the system are susceptible to CPU throttling under certain conditions either to conserve power or to reduce the amount of heat generated by the chip. In other words, based on the load on the system, the frequency of a microprocessor can be automatically adjusted on the fly. This is referred as "CPU dynamic voltage and frequency scaling" (DVFS). Monitoring the output of powertop is one way to monitor the frequency levels of the processor on a busy system in order to minimize any performance related surprises. Set the power management policy to performance, if letting CPUs run at full speed all the time is desired. Performance policy effectively disables the CPU power management.

Power management settings can be controlled from the Service Processor's (SP) Integrated Lights Out Manager (ILOM) command line interface or browser user interface.

The following sample is gathered from an idle SPARC T5-8 server where the CPU power management was disabled.

                                                    Solaris PowerTOP version 1.3

Idle Power States       Avg     Residency             	Frequency Levels
C0 (cpu running)                (0.1%)                	 500 Mhz        0.0%
C1                      4.7ms   (99.9%)               	 800 Mhz        0.0%
                                                      	 933 Mhz        0.0%
                                                      	1067 Mhz        0.0%
                                                      	1200 Mhz        0.0%
                                                	3200 Mhz        0.0%
                                                	3333 Mhz        0.0%
                                                	3467 Mhz        0.0%
                                                	3600 Mhz      100.0%

Wakeups-from-idle per second: 109818.7  interval: 5.0s
no power usage estimate available

Top causes for wakeups:
94.4% (103630.7)               sched :  <xcalls> unix`dtrace_sync_func
 3.1% (3352.8)              OPMNPing :  <xcalls> unix`setsoftint_tl1
 1.1% (1155.6)                 sched :  <xcalls> unix`setsoftint_tl1
 0.4% (401.2)               <kernel> :  genunix`pm_timer
 0.3% (317.0)                  sched :  <xcalls> 
 0.2% (251.8)               <kernel> :  genunix`lwp_timer_timeout
 0.2% (204.4)                  sched :  <xcalls> unix`null_xcall
 0.1% (100.2)               <kernel> :  genunix`clock
 0.1% ( 65.6)               <kernel> :  genunix`cv_wakeup
 0.0% ( 50.2)               <kernel> :  SDC`sysdc_update
 0.0% ( 46.8)            <interrupt> :  mcxnex#0 
 0.0% ( 39.6)                   opmn :  <xcalls> unix`setsoftint_tl1
 0.0% ( 36.6)                   opmn :  <xcalls> 
 0.0% ( 36.4)                   opmn :  <xcalls> unix`vtag_flushrange_group_tl1
 0.0% ( 21.6)            <interrupt> :  ixgbe#0

Suggestion: enable CPU power management using poweradm(1m)

Q - Quit R - Refresh (CPU PM is disabled)

Saturday Dec 21, 2013

Measuring Network Bandwidth Using iperf

iperf is a simple, open source tool to measure the network bandwidth. It can test TCP or UDP throughput. Tools like iperf are useful to check the performance of a network real quick, by comparing the achieved bandwidth with the expectation. The example in this blog post is from a Solaris system, but the instructions and testing methodology are applicable on all supported platforms including Linux.

Download the source code from iperf's home page, and build the iperf binary. Those running Solaris 10 or later, can download the pre-built binary (file size: 245K) from this location to give it a quick try (right click and "Save Link As .." or similar option).

Testing methodology:

iperf's network performance measurements are based on the client-server communication model - hence requires establishing both a server and a client. The same iperf binary can be used to run the process in server and client modes.

  1. Start iperf in server mode
    iperf -s -i <interval>

    Option -s or --server starts the process in server mode. -i or --interval is the sampling interval in seconds.

  2. Start iperf in client mode, and test the network connection between client and the server with arbitrary data transfers.

    iperf -n <bytes> -i <interval> -c <ServerIP>

    Option -c or --client starts the process in client mode. Option -n or --bytes specify the number of bytes to transmit in bytes, KB (use suffix K) or MB (use suffix M). -i or --interval is the sampling interval in seconds. The last option is the IP address or the hostname of the server to connect to. By default, client connects to the server using TCP. -u or --udp switches to UDP.

  3. Check the network link speed on server and client, and compare the throughput achieved.

Check the man page out for the full list of options supported by iperf in client and server modes.

Here is a simple demonstration.

On server node:

iperfserv% dladm show-phys net0
LINK              MEDIA                STATE      SPEED  DUPLEX    DEVICE
net0              Ethernet             up         1000   full      igb0

iperfserv% ifconfig net0 | grep inet
        inet netmask ffffff00 broadcast

iperfserv% ./iperf -v
iperf version 3.0-BETA5 (28 March 2013)SunOS iperfserv 5.11 11.1 sun4v sparc sun4v

iperfserv% ./iperf -s -i 1
Server listening on 5201

On client node:

client% dladm show-phys net0
LINK              MEDIA                STATE      SPEED  DUPLEX    DEVICE
net0              Ethernet             up         1000   full      igb0

client% ifconfig net0 | grep inet
        inet netmask ffffff00 broadcast

client% ./iperf  -n 1024M  -i 1 -c
Connecting to host, port 5201
[  4] local port 63507 connected to port 5201
[ ID] Interval           Transfer     Bandwidth
[  4]   0.00-1.01   sec   105 MBytes   875 Mbits/sec
[  4]   1.01-2.02   sec   112 MBytes   934 Mbits/sec
[  4]   2.02-3.00   sec   110 MBytes   934 Mbits/sec
[  4]   8.02-9.01   sec   110 MBytes   933 Mbits/sec
[  4]   9.01-9.27   sec  30.0 MBytes   934 Mbits/sec
[ ID] Interval           Transfer     Bandwidth
[  4]   0.00-9.27   sec  1.00 GBytes   927 Mbits/sec
[  4]   0.00-9.27   sec  1.00 GBytes   927 Mbits/sec

iperf Done.

At the same time, somewhat similar messages are written to stdout on the server node.

iperfserv% ./iperf  -s -i 1
Server listening on 5201
Accepted connection from, port 33457
[  5] local port 5201 connected to port 63507
[ ID] Interval           Transfer     Bandwidth
[  5]   0.00-1.00   sec   104 MBytes   874 Mbits/sec
[  5]   1.00-2.00   sec   111 MBytes   934 Mbits/sec
[  5]   2.00-3.00   sec   111 MBytes   934 Mbits/sec
[ ID] Interval           Transfer     Bandwidth
[  5]   0.00-9.28   sec  1.00 GBytes   927 Mbits/sec
[  5]   0.00-9.28   sec  1.00 GBytes   927 Mbits/sec
Server listening on 5201

The link speed is specified in Mbps (megabit per second). In the above example, the network link is operating at 1000 Mbps speed, and the achieved bandwidth is 927 Mbps, which is 92.7% of the advertised bandwidth.


  • It is not necessary to execute iperf in client and server modes as root or privileged user
  • In server mode, iperf uses port 5201 by default. It can be changed to something else using -p or --port option
  • Restart iperf server after each client test to get reliable, consistent results
  • Using iperf is just one of many ways to measure the network bandwidth. There are other tools such as uperf, ttcp, netperf, bwping, udpmon, tcpmon, .. just to name a few. Research and pick the one that best suits your requirement.

Saturday Nov 30, 2013

Things to Consider when Planning the Redo logs for Oracle Database

Very basic and generic discussion from the performance point of view. Customers still have to do their due diligence in understanding redo logs, and how they work in Oracle database, before finalizing redo log configuration for their deployments.

  • size them properly
    • log writer writes to a single redo log file until either it is full or a manual log switch is requested
          Oracle supports multiplexed redo logs for availability, but this behavior of writing to a file until it is full or a log switch happens, still hold
    • if the transactions generate a lot of redo before a database commit, consider large sizes in tens of gigabytes for redo logs
    • if not sized properly, it leads to unnecessary log switches, which in turn increase checkpoint activity resulting in unnecessary slow down of the database operations
          two redo logs each with at least 5G in size might be a good start. observe the log switches, checkpoints and increase (or decrease, though there is no performance benefit) the file size accordingly

  • do not mix redo logs with the rest of the database or anything else
    • in a normal functioning database, most of the time, log writer simply writes redo entries sequentially to redo logs
    • any slow down in writing the redo data to logs hurt the performance of the database
    • best not to share the disks/volumes on which redo logs are hosted, with anything else
          set of disks, volumes exclusive to redo logs, that is

  • ensure that the underlying disks or I/O medium used to store the redo logs are fast, optimally configured and can sustain the amount of I/O bandwidth needed to write the redo entries to the redo logs
        if those requirements are not met, it could lead to 'log file sync' waits, which will slow down the database transactions

  • redo logs on non-volatile flash storage may have performance benefits over the traditional hard disk drives
    • check this blog post out, Redo logs on F40 PCIe Cards, for related discussion (keywords: 4K block size for redo logs, block alignment)

Benchmark announcements, HOW-TOs, Tips and Troubleshooting


« October 2015