BM Seer Unofficial thoughts from an anonymous Sun employee

« Burning Wild - with... | Main | Watts from IBM,... »

Why Solaris? performance is a key factor in why

Tuesday Sep 19, 2006

This afternoon, I got asked a couple of times for quantitative comparisons of Solaris vs. Linux on the exact same hardware. Below is one comparison I've mentioned before. I'll also hopefully post another next week.

SPECjbb2005 on Sun Fire X4200 4-core, we see a Solaris10 score of 49,097 bops and on RedHat Linux we get a score of 43,076 bops -- net, net this makes Solaris10 is 14% Faster than Linux! It is important to note that SPECjbb2005 is not known as a heavy OS benchmark, and still we get this performance improvement.

On Windows2003 we see a score of 47,437 bops, this makes Solaris10 is 3% Faster than Windows. For more information on this benchmark, go to the SPEC site listed in the disclosure at the very end.

A blog entry listed below points to a pretty long article comparing Solaris to Linux (sorry, I've only had time to skim the article) http://blogs.sun.com/nealix/entry/solaris_10_vs_linux

Disclosure Statement

SPECjbb2005 Sun Fire X4100/X4200 (2 chips, 4 cores, Solaris 10) 49,097 bops, 49,097 bops/JVM, Sun Fire X4100/X4200 (2 chips, 4 cores, Windows2003) 47,437 bops, 47,437 bops/JVM, Sun Fire X4100/X4200 (2 chips, 4 cores, RHEL4u2) 43,076 bops, 43,076 bops/JVM. SPEC, SPECjbb reg tm of Standard Performance Evaluation Corporation. See www.spec.org 1/27/06.

[7] Comments

Like this post?

Posted at 05:05PM Sep 19, 2006 in Competitor Follies
Tags:

Comments:

Since when is one benchmark, on one piece of hardware, indicative of overall OS performance? This seems like the same rationale Microsft uses...

Posted by Alex Valentine on September 19, 2006 at 07:04 PM PDT #

Seems the options to the JVM weren't the same across the three results? "Linux" used XX:+UseLargePage where Solaris did not. To what extent did that matter? Is there a difference in the supported page sizes between the two OSes? Similarly, the options on Windows were different "-Xmn750m -Xms1500m -Xmx1500m " vs "Xmn2g -Xms3g -Xmx3g" for Solaris. Does that mean that Windows was doing its thing with less than half the memory? How about the slightly later Windows Tyan result using What appear to be the same Opteron 285's and getting 51107? http://www.spec.org/osg/jbb2005/results/res2006q1/jbb2005-20060228-00086.html

Posted by rick jones on September 19, 2006 at 08:04 PM PDT #

Its actually more than just one benchmark, Alex. One area close to home, SPEC FP and SPEC OMP, SunStudio+Solaris beats Linux with ANY compiler. You can argue that when SunStudio releases on Linux, will this change the picture... we'll know when all the i's are dotted and t's crossed. But for now, thats an advantage, and in HPC world, its a good lead. Presumably these URLs would help:
http://blogs.sun.com/tatkar/entry/highest_spec_fp_on_the
http://www.sun.com/servers/x64/x4600/benchmarks.jsp#question4
http://www.sun.com/servers/x64/x4600/benchmarks.jsp#question3

Posted by 192.18.43.249 on September 20, 2006 at 01:37 PM PDT #

I don't see that any of these benchmarks prove Solaris is better. In all cases the application software (jvm or compiler) is also critical to the performance of the benchmark. Even if SunStudio is released on Linux, what guarantees are there that the optimizations will be as good? Sun may intentionally handicap it to make Solaris look better!

Posted by BM Xrayer on September 20, 2006 at 08:06 PM PDT #

Rick, Xray,

The point of the benchmarks was to use the exact same hardware and exact same JVM (down to the dot-rev). That is as close a comparison which can be made. Yes Rick, the Tyan result, posted three months later than the Sun result, and using a later JVM, is higher. The Windows OSs are different as well. 64-bit on Sun vs. 32-bit SP1 on the Tyan. There is one difference in the JVM parameters. There are some differences in the OS parameters. All of these may represent minor tuning tweaks which allowed Tyan to get a better score. Good on them. Certainly Sun's JVM probably works better on Solaris than other OSs. Certainly Solaris is more optimized for Java than other OSs. But none of these facts detract from the value of the Sun trifecta.

The days of the raw speed a 3 GHz Pentium processor being a reason to dump Solaris for Linux are gone. The days of a "free" OS being a reason to dump Solaris for Linux are gone.

At the same time, Sun offers solid x86 systems. The days of IBM xSeries or HP Proliants being an excuse to dump Sun servers are gone. And the days of Sun not running Linux or Windows is gone.

But the really cool thing is the return of Solaris to its roots as the premier distributed computing OS. Solaris, both on x86 and UltraSPARC T1 processors, seems the perfect platform for horizontally scaled application servers.

Posted by Mark on September 20, 2006 at 09:55 PM PDT #

Hey, even Google is thinking about OpenSolaris (and those guys always want top perf and lowest $. http://www.computerworld.com/action/article.do?command=viewArticleBasic&articleId=9003492

Posted by BM Seer on September 21, 2006 at 03:21 PM PDT #

Whoops forgot about this test as well.... http://www.sun.com/smi/Press/sunflash/2006-04/sunflash.20060421.2.xml

Posted by BM Seer on September 21, 2006 at 03:25 PM PDT #