Thursday, December 22, 2005

Solaris 10 Update 1

Solaris 10 Update 1 is available. There're many changes comparing to S10 03/05 like:

  • New Boot Architecture on x86/x64 (GRUB, faster system startups, easier network installs,...)
  • iSCSI client support
  • GLDv3 (VLANs, Link Aggregation, improved performance - man dladm)
  • Performance improvements (MPO, Large Pages, ...)
  • fcinfo command
  • new drivers
  • new ACPI (+virtual keyboard&mouse)
Check What's New.

ps. of course it is available for free

Wednesday, December 21, 2005

Writing Zeros to ZFS files

I did look into some of the ZFS code and found nice little feature. In a function zio_compress_data there is:
75 /*

76 * If the data is all zeroes, we don't even need to allocate

77 * a block for it. We indicate this by setting *destsizep = 0.

78 */

79 allzero = 1;

80 word = src;

81 word_end = (uint64_t *)(uintptr_t)((uintptr_t)word + srcsize);

82 while (word < word_end) {

83 if (*word++ != 0) {

84 allzero = 0;

85 break;

86 }

87 }

88 if (allzero) {

89 *destp = NULL;

90 *destsizep = 0;

91 *destbufsizep = 0;

92 return (1);

93 }



So if you set compression=on (or set any other available compression method - right now there's only lzjb available) if a block contains all zeros then actual compression method will not be called (less CPU consumed) and no IOs for these zeros will be generated. I did a small test - created two filesystems on a RAID-Z pool - one with compression set to on and other to off. Then I did run 'dd if=/dev/zero of=/test/fs1/q1 bs=1024k count=1024' which translates to writing 1GB of data (only zeros). If a filesystem is with compression set to off then using iostat I can see lot of IOs to underlying disks. However if compression is set to on then only about 600KB is written to underlying disks (not to mention that whole operation took about 2s on a 1.5GHz USIIIi). Well, this is really clever small thing (not the most clever in ZFS of course).

Now I wonder if it would be beneficial if the same check would be done even if there's no compression? (basically moving this code up - so regardless if compression is on or off if whole block is 0s then do now write them).

LDAP hints

If you are interested in Directory Server and some hints, tuning information then you should look at this blog. You can find them some info why using ZFS with LDAP is a good idea.


btw: if you haven't known already - Directory Server is free to use with no limits as part of Solaris Enterprise System.

Wednesday, December 14, 2005

Linux Applications on Solaris

Want to run Linux applications (and whole CentOS or RedHat) in a Zone on a Solaris? If you do please check its official page. Well, even DTrace works with Linux applications.

I did test JANUS (it was called Project JANUS) some time ago and basically it just worked - mozilla, yum, other applications. I did try with CentOS.

From the announcement:

BrandZ is a technology that extends the zones infrastructure to allow for the creation of "non-native" zones. Non-native is a deliberately ambiguous term, as we are trying not to let our preconceived notions of the technology limit its usefulness.

The first brand we are developing under the BrandZ umbrella is 'lx', a brand that supports the execution of 32-bit x86 Linux applications on a x86/x64 machine running Solaris Nevada. Specifically, the lx brand allows the user to install a complete CentOS or Red Hat Enterprise Linux 3.x distribution in a zone. When the zone is booted it will still be running the Solaris kernel under the hood, but the userspace environment will include nothing but Linux software from init(1M) on up.

In theory the BrandZ infrastructure could also be used to create other types of zones. One such example would be a GNU Solaris brand, which runs Solaris binaries but has the standard utilities replaced by their GNU equivalents. Other possible uses would be the creation of zones for running FreeBSD or Darwin x86 environments. Because this technology is being made available via the OpenSolaris community, you as a community member will be able to help create these or other brands if so inspired.



If you are interested in a background information about BrandZ read it.

Monday, December 12, 2005

T1000/T2000 - Benchmarks Records

Systemnew reports:

* Sun Fire T2000 server using Sun JavaTM System Web Server 6.1 SP5 achieved a world record SPECweb2005 and demonstrated 1.7 times the performance advantage over the four-way IBM eServer p5 550 with 4.3 times higher performance per watt while occupying half the space. In addition, the Sun Fire T2000 server delivers more than three times the performance of the two-way 3.8 GHz Xeon-based IBM eServer xSeries x346 while delivering over 4.1 times higher performance per watt.

* Sun Fire T2000 server using BEA Weblogic Server was 1.3 times faster than the performance of a four-way 1.6 GHz Itanium two-based HP rx4600 server on the dual node SPECjAppServer2004. The Sun Fire T2000 achieved the overall performance world record on all two node results.

* Sun Fire T2000 server using Sun JavaTM System Application Server 8.2 Performance Edition achieved world record price/performance on the application tier beating a four-way 1.6GHz Itanium two-based HP rx4600 server on the dual node SPECjAppServer2004.

* Sun Fire T2000 server, equipped with the UltraSPARC T1 processor achieved overall price/performance leadership on the Lotus R6iNotes Domino 7 benchmark. On IBM's own benchmark the Sun Fire T2000 beats the price/performance of the POWER5+ based IBM p5 p550Q by 27 percent. In addition, Sun has more than twice the price/performance advantage and a nine percent performance advantage over the POWER5 based IBM p5 570 server.

* In a demonstration addressing portal server workload, the Sun Fire T2000 server beats the performance of the 2GHz Xeon-based Dell 6650 server by running on the new Sun JavaTM System Portal Server 7, with 6 times more logins per second while providing 33 percent capacity headroom on Sun Fire T2000 Server versus zero percent on Dell. This new release of the Java System Portal Server allows users to easily create interactive communities of users and services, building "community" portals populated with collaborative content including RSS feeds, blogs and wikis. Additional information on the Java System Portal Server 7 is expected from Sun in the near future.

* Sun Fire T1000 beats the performance of the Dell SC1425 by over two times, while consuming half the power. In comparison to the IBM p520 two-way Power 5+ server, the Sun Fire T1000 server delivered 1.5 times higher performance in four times less space and at 3.7 times performance per watt.

* Sun Fire T2000 server beats the performance of the 1.9 GHz POWER5+ based IBM p5 550 four-way on the SPECjbb2005 and was more than 1.6 times faster than the 2.8 GHz dual-core IBM x346. HP and Fujitsu have not published results on this new benchmark that demonstrates JavaTM server performance.

Friday, December 09, 2005

Measuring NFS traffic with DTrace

Below you can find simple script written in DTrace which will show you in a given time intervals what is a distribution of data written and read to/from NFS mounted filesystems in Mb per second (average in the interval). It will be updated continuously every given amount of time.

Script expects four arguments: low high step interval. First two of them are low and high of a range in Mbs you want your output, step describes resolution of the output and interval is a resolution of a measurement (time interval from which Mbs is calculated). In an example below I wanted to see output in a range 50-150Mbs with 5Mbs step calculated for intervals of 10s and see results every 10s. As you can see most of the time I have on average 90-120Mbs generated to/from NFS servers.
Note that this is only an estimate.


bash-3.00# cat io-nfs-throughput.d
#!/usr/sbin/dtrace -qs

BEGIN
{
bsum = 0;
}

io:nfs::start
{
bsum += args[0]->b_bcount;
}

tick-$4
{
@a["Mbs"] = lquantize(bsum*8/($4*1024*1024),$1,$2,$3);
printa(@a);
bsum=0;
}

bash-3.00# ./io-nfs-throughput.d 50 150 5 10s
[...] after some time
Mbs
value ------------- Distribution ------------- count
55 | 0
60 | 1
65 | 2
70 | 2
75 |@ 3
80 |@ 8
85 |@@@ 17
90 |@@@@@ 32
95 |@@@ 16
100 |@@@@ 23
105 |@@@@ 22
110 |@@@@ 21
115 |@@@@ 21
120 |@@@ 19
125 |@@@ 17
130 |@@ 14
135 |@@ 11
140 |@ 7
145 | 0
>= 150 |@ 4

Simulating Data Center

This is cool simulator of data center :))) You can easily compare T1000/T2000 with IBM's POWER5 servers (or other server if you define them your self). Not only performance but also power usage, space usage and costs. Have a great fun!

ZFS data integrity

A must read.

Thursday, December 08, 2005

snv_b28 available

Solaris Express snv_b28 is available both CD ISOs and DVD ISOs. Some interesting changes (comparing to b27) from putback logs:
  • 64-bit glm driver on x86
  • Interactive volume creation and deletion for raidctl
  • several ZFS bug-fixes and performance improvements
  • Kernel-level SSL proxy
I haven't blogged it yet (I was on vacations) but starting from build 27 ZFS is available to public! Also sources of ZFS are available at opensolaris.org. So if you want to play with ZFS (of course you want!) go and download SX (of course it's also available in latest bits - b28). Other OpenSolaris based distributions seems to already include ZFS too (http://www.gnusolaris.org). If you have any questions about ZFS go and subscribe to ZFS mailing list.

Wednesday, December 07, 2005

Details about Ultra Sparc T1

These are a must read!

CMT Scheduling Optimizations
Vertical Multithreading
Niagara - Network Throughput

Niagara - Crypto Acceleration
Niagara - low power and SWaP metric
Niagara - large pages
Niagara - I/O Performance
T2000 - RAS
Several other links about T1 by Richard McDougall
RSA Performance of Sun Fire T2000


Network Computing O5Q4

During yesterday's Network Computing Sun announced new servers: T1000 and T2000. Both with Niagara processors (16-32 threads per CPU). Below some Copy&Paste from announcements.

http://www.sun.com/smi/Press/sunflash/2005-12/sunflash.20051206.3.html
Continuing to build on their longstanding collaboration, today Sun and Oracle are also offering customers a special opportunity to try Oracle on the Sun Fire T1000 and T2000 systems. As part of a special promotion, customers using Oracle products with CPU-based licenses on Sun Fire T1000 and T2000 systems will be able to count cores as .25 percent of a processor versus alternative methods.
http://www.sun.com/smi/Press/sunflash/2005-12/sunflash.20051206.4.html
Today, Sun also announced plans to publish specifications for the UltraSPARC-based chip, including the source of the design expressed in Verilog, a verification suite and simulation models, instruction set architecture specification (UltraSPARC Architecture 2005) and a Solaris OS port. The goal is to enable community members to build on proven technology at a markedly lower cost and to innovate freely. The source code will be released under an Open Source Initiative (OSI)-approved open source license.
OpenSparc Project Page


http://www.sun.com/smi/Press/sunflash/2005-12/sunflash.20051206.2.html

Sun Fire T2000 server using Sun's Web Server 6.1 SP5 achieved a world record SPECweb2005 and demonstrated a 1.7x performance advantage over the 4-way IBM eServer p5 550 with 4.3x higher performance per watt while occupying half the space. In addition, the Sun Fire T2000 delivers more than 3x the performance of the 2-way 3.8GHz Xeon-based IBM eServer xSeries x346 while delivering 4.1X higher performance per watt.(1)

Sun Fire T2000 server using BEA Weblogic Server was 1.3X faster than the performance of a 4-way 1.6GHz Itanium2-based HP rx4600 server on the dual node SPECjAppServer2004. The Sun Fire T2000 achieved the overall performance world record on all two node results.(2)

Sun Fire T2000 server using Sun Java System Application Server 8.2 Performance Edition (AS 8.2 PE) achieved world record price/performance on the application tier beating a 4-way 1.6GHz Itanium2-based HP rx4600 server on the dual node SPECjAppServer2004. Sun Java System AS 8.2PE is free for development and deployment.

Sun Fire T2000 server, equipped with the UltraSPARC T1 processor achieved overall price/performance leadership on the Lotus R6iNotes Domino 7 benchmark. On IBM's own benchmark the Sun Fire T2000 beats the price/performance of the POWER5+ based IBM p5 p550Q by 27 percent. In addition, Sun has more than twice the price/performance advantage and a nine percent performance advantage over the POWER5 based IBM p5 570 server. (3)

In a demonstration addressing the all-important and ubiquitous portal server workload, the Sun Fire T2000 server beats the performance of the 2GHz Xeon-based Dell 6650 server by running on the new Sun Java System Portal Server 7, with 6x more logins per second while providing 33 percent capacity headroom on Sun Fire T2000 Server versus zero percent on Dell. This new release of the Sun Java System Portal Server allows users to easily create interactive communities of users and services, building "community" portals populated with collaborative content including RSS feeds, Blogs and Wikis. Additional information on the Sun Java System Portal Server 7 will be communicated in the coming days.

Sun Fire T1000 beats the performance of the Dell SC1425 by over 2x, while consuming half the power. In comparison to the IBM p520 2-way Power 5+ server, the Sun Fire T1000 server delivered 1.5x higher performance in 4x less space and at 3.7x superior performance per watt.

Sun Fire T2000 server beats the performance of the 1.9GHz POWER5+ based IBM p5 550 4-way on the SPECjbb2005 and was more than 1.6x faster than the 2.8GHz dual-core IBM x346. (4) Surprisingly, HP and Fujitsu have not published results on this important new benchmark that demonstrates Java server performance.

Tuesday, December 06, 2005

T2000

SystemNews is reporting new T2000 SPARC server from Sun based on Niagara chip UltraSPARC T1. You can find more info also on Sun pages. Looks like this server is going to be presented on tomorrow NC event. Chassis looks similar to x4200 - probably the same. From the documentation:

  • CPU - 4, 6, 8 cores (each with 4 threads, so 16, 24 or 32 virtual CPUs are seen in a system) UltraSPARC T1 (Niagara)
  • RAM - 16 slots (with 2GB DDR2 - 32GB)
  • Ethernet - 4x 100/1000 on-board
  • DISKS - 1-4 2,5" SFF SAS 73GB
  • 1x DVD
  • I/O - 3x PCI-E, 2x PCI-X (64bit 133MHz)
  • Redundant power and cooling
  • OBP/ALOM
  • Hardware-assisted cryptography (RSA & DSA on-chip)
  • 350W nominal power consumption (400W MAX)

Now I wonder about price of the server and what actual performance in web serving it could achieve.

btw: looks like smaller version named T1000 is going to presented too.

update: another article on Niagara with some performance benchmarks.
From the article:

• On the SPECjbb2005 test of Java server software, the T2000 scored 53,378 business operations per second compared with 61,789 for an IBM p5-550 with two dual-core Power5 chips and 24,208 for a Dell PowerEdge SC1425 with dual single-core Xeon processors.

• On the SPECweb2005 test of Web server performance, the T2000 socred 14,001, compared with 7,881 for an IBM p5-550 with two dual-core Power5 processors, 4,850 for a Dell PowerEdge 2850 with two dual-core Xeon processors, and 4,348 for an IBM x345 with dual single-core Xeon processors.

• On the NotesBench test of Lotus Notes performance, a T2000 accommodated 19,000 users at $4.35 per user and got a NotesMark score of 16,061. In comparison, an eight-processor IBM p5-570 had 17,400 users, a cost of $10.19 per user, and a NotesMark score of 14,740. But the average response time of the IBM system was 270 microseconds compared with the slower 400 microseconds for the T2000, demonstrating the relatively slow single-thread performance of the Sun system.


Looks like WEB performance is really good.

Monday, December 05, 2005

NexentaOS


I've just installed NexentaOS on my laptop - this is GNU distribution based on Debian and Open Solaris. I must admit that it works really good - automatically detects my network, X, etc. - quite impressive. Looks like DTrace, ZFS, SMF works just out of the box. Tools for packet administration also work - I tried to upgrade some of packages and it just worked! Good job. I think this could become great distribution for all people who prefer GNU utilities but want Solaris stability and features like DTrace, ZFS, etc. It's definitelly worth to look at it. Debian users will probably feel quite comfortable with this distribution. Keep in mind that this is alpha release (still I'm impressed how much has been achieved in such a short time - Open Solaris went public in June this year).

Solaris Networking internals

Part I and Part II of Solaris Networking - the Magic Revealed is going to be a part of new (upcoming) Solaris Internals book.

Friday, December 02, 2005

Project Red October

Project Red October - Sun is giving JES, N1 and developer tools for free! And it looks like all of the software will be open sourced. It means that for example Sun Cluster or Messaging Server, C/C++ compilers, etc. are now free on all available platforms.

From official announcement:
MENLO PARK, CALIF; November 30, 2005 - Sun Microsystems, Inc. (NASDAQ: SUNW) today announced two landmark moves in the battle to create the software platform of choice for the next-generation of the Internet. First, having seen tremendous momentum with the Solaris Operating System (OS) as free and open source software, Sun is making the Java Enterprise System, Sun N1 Management software, and Sun developer tools available at no cost for both development and deployment and further, is reaffirming its commitment to open source this software. Second, Sun is announcing that it is integrating all of this software along with the Solaris OS into the Solaris Enterprise System, the only comprehensive and open infrastructure software platform available today.
All of Sun's server software is delivered in one pack and is called Solaris Enterprise System.