Benchmarking and systems performance - hosted by Elisabeth Stahl

Elisabeth Stahl

Autores do Blog:  ELISABETH STAHL   is Chief Technical Strategist and Executive IT Specialist, IBM Systems and Technology Group, and has been working in systems performance for over 25 years.

Comentários (4)

1 Alex Fatkulin comentou às Link permanente

"...not the 6-core one their press release leads you to believe"<div>&nbsp;</div> The press release says: "with eight, Six-Core AMD Opteron™ 8439 SE 2.8 GHz processors". This is clear about eight six core CPUs being used. 8x6 = 48. I don't see any confusion here?<div>&nbsp;</div> "Note that while Oracle reports 7x the performance of IBM it was with 12x the number of processor cores".<div>&nbsp;</div> What's the deal in comparing Power6 at 4.7GHz with AMD Opteron at 2.8GHz after all? This is meaningless. What is not meaningless is response time, throughput, etc. figures. Whether these archived on smaller amount of faster cores or on greater amount of slower cores, who cares? There are different ways to perform, you know.

2 estahl comentou às Link permanente

Thanks to Oracle for changing their press release after it was issued to specify "eight, Six-core."

3 Andrew_Larmour comentou às Link de Retorno

@Alex, regardless of why Oracle chose to compare the RISC and CISC platforms, the result is not meaningless. Java App Servers are licensed on cores that they run on. For most companies, bang for buck is a key measure. As the comparison is JEE compliant Java App Servers, we can legitimately compare app servers running on different underlying hardware architectures and come up with a legitimate measure of bang for buck. <br /> . <br /> Let's look at IBM's and Oracle's licensing for this benchmark; the WebLogic instance would require 24 CPU equivalent*** (48 x 0.5) licenses of Weblogic App Server (Enterprise Edition) while the Power system will require 480 Processor Value Units (4x120) of WebSphere Application Server Network Deployment. To compare the two models if we take the list price for WebLogic Enterprise Edition (US$30,500* per CPU equivalent) and IBM's WebSphere App Server ND (US$174/PVU*) then we see that <strong>the Oracle WebLogic App Server license (US$732,000) will cost 776% more than the IBM WebSphere App Server license (US$83,520).</strong> <br /> . <br /> Oracle are claiming "...nearly 7 times..." the performance despite the fact that 9,455.17/1,197.51 = 7.90 (to 2 decimal places) which in my book is nearly 8 times the performance, not nearly 7 times. I think their marketing people got their percentages mixed up - 7.90 times the performance of the IBM score is a 690% improvement on the IBM score. <br /> . <br /> So, let's give them the benefit of the doubt and assume the Oracle marketing folks made a mistake and their benchmark system can deliver <strong>7.9 times the performance</strong> of the IBM benchmark system, they are doing it <strong>for 8.7 times the price</strong> of the IBM system in terms of app server licensing, that is not looking like the spectacular win that Oracle are claiming it to be... <strong>In the bang for buck war (at least in software licenses), IBM still wins.</strong> <br /> . <br /> * List price including support &amp; maintainance <br /> ** Prices from Oracle and IBM's web sites - valid at 30Nov09 <br /> *** See my post (<a href="" target="new">Simple explanation of PVU licensing and Oracle's equivlent for multi-core CPUs</a>) for a simple explanation of Oracle and IBM software licensing when it comes to multi-core hardware.

4 Alex Fatkulin comentou às Link permanente

@Andrew_Larmour Plugging in software costs just drives my own point. The post should care about the total "system" cost/performance as opposed to stripping out individual components (how about we compare the price of POWER6 CPU vs AMD Opteron CPU, now what?). But one who says A should say B. <div>&nbsp;</div> Look at the latest TPC-C benchmarks where Oracle beats you with 384 cores vs 64 cores for the IBM system in both performance and price/performance figures. Turns out the lower amount of cores does not necessarily results in a better "bang for the buck"? <div>&nbsp;</div> I guess you have to consider IBM's hardware costs too.

Incluir um Comentário Incluir um Comentário