BM Seer Unofficial thoughts from an anonymous Sun employee

« IBM power6 chip... | Main | Current SPECjbb2005... »

IBM rewrites history, OK footnotes to clearly show bogus calculations

Tuesday Jun 19, 2007

IBM rewrote the press release, now these footnotes validate all of BM Seer's calculations. Yes gang they make lots of very bad assumptions on the bogus 30x Sun comparison.

I don't know when they re-wrote this, I mistakenly clicked on their press release again and noticed they added even more footnotes.

They really should have just struck the whole claim because basically they just applied factors to make a big bogus number. You can read the whole press release at: "IBM Unleashes...".

To summarize:

They used a 2-year old Sun result against a IBM product they just announced

They compared hugely expensive IBM $1M system versus a much cheaper Sun systems.

IBM assumed a conservative 5:1 performance ratio, of course they were conservative by giving themselves another 15%

IBM then assumed a IBM utilisation of 60% versus a Sun utilisation of 20%, which nicely gave them another 3x advantage - Hey gang remember I pointed this out right after launch, but I had to dig threw reams of IBM docs to find it.

Of course then they used max power but never pointed to any docs to back that up - thanks to a comment, we could then find where this was, but it still provides only the vague info on maximum wattage.

I surprised they didn't use Connecticut residential electricity prices to even further make an advantage for more costly electricity.

...all of that gives them a factor of 5*1.15*3 = 17.5x then I'm sure they put in some more factors due to racking? or? I guess that gives them 30x - what a load of

OK so what is the reality, If you take several of the Sun V890 at 2.1GHz I bet it out perform the p570 systems and would still cost losts less than a IBM $1M hairball, but enough of estimates, let's go back to real performance result comparisons. And IBM just delete this bogus comparison. Really it is embarrassing to the industry. Also quit trying to make 300GB/s processor bandwidth sound like system bandwidth that can download itunes library which adds all bandwidth along the memory to CPU pathway together -that doesn't make sense either.

3 (3) This calculation is based on the trend toward consolidation of existing installed systems. Performance comparisons of the system were based on available benchmarks using the Java-based SPECjbb2005 benchmark (results as of 5/22/07: System p 570 (16-core, 8 chips, 2 chips per core, 4.7 GHz) SPECjbb2005 691,975 bops, 86497 bops/JVM; Sun Fire v890 (16-core, 8 chips, 2 chips per core) 1.5 GHz, SPECjbb2005 117,986 bops, 29,497 bops/JVM). A conservative 5 to 1 performance comparison was used. System utilization levels were derived from studies conducted by IBM of currently installed base of UNIX systems (available at www.ibm.com/servers/library/pdf/scorpion.pdf) and the recent trend on System p for utilizations levels well over 60% using advance virtualization technologies. 20% was used for the currently installed base of Sun Fire v890 systems and 60% for a virtualized System p 570. Higher utilization levels provide a 3 to 1 consolidation factor. Power consumption figures of 5600 W for the IBM System p 570 and 3200 W for the Sun Fire v890 were based on the maximum rates published by IBM and Sun Microsystems, respectively. Air conditioning power requirement estimated at 50% of system power requirement. Energy cost of $.092 per kWh is based on 2007 YTD US Average Retail price to commercial customers per US DOE at http://www.eia.doe.gov/cneaf/electricity/epm/table5_6_b.html as of 5/18/2007. For space calculations, two IBM System p 570 servers will fit in a single, standard rack. Assumed rates were 60% for IBM System p 570 and 20% for Sun Fire v890.

[9] Comments

Like this post?

Posted at 11:12AM Jun 19, 2007 in IBM Follies
Tags: ibm power6

Comments:

Noooo, dont use the v890... it is way overpriced for the performance. Plus, unless you need that many internal disks, it is a space and power hog.

Use a Rack of T2000's like I did in an earlier comment. Or for some fun you can use a 6000 series blade server loaded up with the new T6400 blades. A 6000 loaded up with 10 T6300's will run a bit under 200k and provide close to 800,000 bops. It would only take up 10RU vs IBM's 16 (its 2 cpu's per 4U module if i remember correctly)

Posted by John on June 19, 2007 at 02:45 PM PDT #

Interesting calculation, pretty straight-forwared any idea why IBM marketing created such silly desperate calculations like the 30x for power6???

I used the V890 directlly to show their silly reasoning.

Posted by BM Seer on June 19, 2007 at 04:17 PM PDT #

[Trackback] BMSeers points to an rewritten press release of IBM that shows how silly the the "Power6 30x Sun comparision" is and how IBM benchmarketing works.. At least IBM clarified at the end, how those numbers were crafted. My favourite one:IBM then assumed a I...

Posted by c0t0d0s0.org on June 19, 2007 at 11:16 PM PDT #

>They used a 2-year old Sun result against a IBM product they just announced

I see, only you is permitted to compare against 2 years old result in <a ref="http://blogs.sun.com/bmseer/entry/not_comparing_e25k_p595"> E25k vs p595 :)

>They compared hugely expensive IBM $1M system versus a much cheaper Sun systems.

cheaper SUN with TCO 2x against Power6 :)

>If you take several of the Sun V890 at 2.1GHz

If I take several Opteron at 3.0Ghz versus much costly SPARC ... Power6 and SPARC can't win on x86 in 1-4-way market.

Posted by Triffids on June 20, 2007 at 02:24 AM PDT #

You wouldn´t use UltraSPARC IV+ at all for AppServer in most cases. UltraSPARC T1 or Niagara 2 would be the more sensible choice. And those system are competitive against Opteron. When you take into consideration, that you stay on the same processor command set and the same operating system ... that´s invaluable .... Migrating from Solaris on Sparc to AIX on Power6 ... much more work (operating guidelines, other operating system skills, etc)

Posted by Joerg Moellenkamp on June 20, 2007 at 05:44 AM PDT #

2Joerg Moellenkamp
Niagara2 is to slow for App server, where are only 8 slow cores in T2 servers, but it cost $32K. More powerfull 4-way Opteron (8 cores) will cost less than $20K. I don't thinks that Java code migration from SPARC to Linux x86 will require much work and don't believe that SPARC/Solaris admin will be cheaper than Linux x86.

There are no real competitors for x86 on 1-4 way server market, SPARC and Power6 rule on more than 8-way servers ...

Posted by Triffids on June 20, 2007 at 06:16 AM PDT #

Niagara too slow for an app server? That may or may not be true. It all depends on how much processing the application needs to do.

Most of the time, the actual processing per request is low. The app server is just making a bunch of requests for data, formatting it and spitting it back to the user. Thus the response time is 10% processing and 90% waiting for data from things like a database server. In these cases, the Niagara works well since it can deal with 32+ of these at the same time without increasing response time. Other systems may be able to cut the processing time in half, but are still at the mercy of the external systems for the bulk of the response time.

We are upgrading our OneWorld/JDE/peoplesoft/Oracle environment from the old fat client based solution to a web based system. The app server? 1.2GHz T2000's. There was no response time advantage with putting it on a "faster" system. We would have just bumped up the power requirements while providing lower overall throughput. (we also got them cheap as part of the 25th anniversary special a few weeks back)

Posted by John on June 20, 2007 at 03:41 PM PDT #

T200 SPECjbb2005 bops = 74365
http://www.spec.org/osg/jbb2005/results/res2006q2/jbb2005-20060512-00116.html

Xeon (2 chips, 8 cores) SPECjbb2005 bops = 220648
http://www.spec.org/osg/jbb2005/results/res2007q1/jbb2005-20070213-00259.html

This xeon is 4x cheaper, 33% faster and provide 16 threads, if you need 32 threads - get 4-way Opteron, it still cheaper than T200 :)

Posted by Triffids on June 20, 2007 at 11:49 PM PDT #

Sorry, 2-way Xeon 77% faster T200 :)

Posted by Triffids on June 21, 2007 at 12:28 AM PDT #