[Editor: I know this is largely an advertisement for Windows Server, but some of the benchmark data looks interesting, and you may wish to look at it and debate it.] Just how does a high-end IBM RISC system with Power6 chips and WebSphere 7 compare to running IBM WebSphere 7 on a typical Windows Server 2008 configuration in both raw performance and cost? See: http://download.microsoft.com/download/4/8/6/486B4B4F-5A87-4B5C-BEEC-455290F83274/IBMPower570_WebSphere_7_%20NET_Benchmark_WinSrv2008.pdf As usual, such benchmarks come with lots of debate. .NET results and a Capacity Planner tool, that enables testing of the same workloads on WebLogic, Oracle OC4J, WebSphere 7 and .NET is available at: http://msdn.microsoft.com/stocktrader

Greg - Good to see that Microsoft still makes enough money off of its Windows and Office monopolies to subsidize your trolling ;-) Peace, Cameron Purdy Oracle Coherence: Data Grid for Java, .NET and C++

Please Cameron, this is below your standard. Do you really need to derail the discussion? Why?

Please Cameron, this is below your standard. Do you really need to derail the discussion? Why?

How can you derail something that is not moving and has no rails let alone direction. I suggest you look up Greg's past posts.

Please Cameron, this is below your standard. Do you really need to derail the discussion? Why?

You are probably right. I did have that inkling of a feeling that I shouldn't press submit, but .. As background, Greg's group at Microsoft once paid the parent company of TSS to write a study that showed that .NET was a zillion times faster and less expensive than J2EE, and then had it published on TSS without any information that linked it back to Microsoft. Peace, Cameron Purdy Oracle Coherence: Data Grid for Java, .NET and C++

..and performance tuning statistics. :) Websphere loves windows so much that it has to make java seem like the usual mess that the Microsoft bunch expect. Want the pain of microsoft instead of the convenience of java: try websphere. Nathan

Some time ago (2004?) there was Sun report about Java performance on Windows and Solaris. Java on Windows was significantly faster. Reporter, Sun employee, concluded that there are much to be done. Seems like AIX has same problems. Sorry, but I forget exact report location.

The reported test environment (per the PDF) was . . . IBM Power 570 (based on Power6 Architecture) running IBM AIX 5.3 Hewlett-Packard C7000 Blade System with 4 HP ProLiant BL460c blades POINT > - Did you have 4 dual core CPU's [vs. 32 Intel Cores - across 4 blades]? Or did you have 8 Dual Core CPUs - each at 4.2 GHz [16 vs. 32]? This isn't clear and either way we already have a mismatch. POINT > - each blade (x4) had 32gig of RAM. The AIX stood solo with 32GIG. POINT > - Where is a Red Hat Linux comparison running on the same HP Blade configuration. I would be REALLY interested to see that. That's more apples to apples.

WAS7 + with 2 GB of Heap ( What's the use of 32 GB RAM in P570) + with 50 web container threads( this influences the concurrent request served and thus TPS). Thread limit(500) at IBM Http Server is useless with the limited capability of web conatiner. Windows with .NET + can use the whole 32 GB of RAM available. + IIS with 400 threads ( I am not an IIS geek, but i guess this enabled the IIS to server 400 concurrent client), The servers are big, but the resources allocated to Websphere is too low Tamil

No, thread settings are not biased. Remember CPU utilization is near 100%; the first obvious sign of too few threads is non-saturated CPU utilization, no matter how many concurrent client requests are made. Also remember two instances of WAS (each with 50 web container threads and separate DB JDBC connection pools are running). We tested different heap sizes, different thread settings, etc...this was not done in a hurry. As with all my tests in the past and this one, I am quite confident in the results and have no desire to mislead anyone. As usual, we have offered to meet IBM in an independent lab, but no response. As for Linux, I have run and published these tests on WebSphere 6.1 on Linux and Windows Server 2008 (different hardware though); you can expect to see roughly the same perf on each platform, at least with WAS 6.1. IBM does an excellent job with their JVM for both platforms; and hence little difference when comparing same hardware. There is a misperception, however, I think, among many, that WebSphere (or WebLogic for that matter) are simply better on Solaris, AIX or Linux vs. Windows. I think also many UNIX/RISC folks somehow believe all the money for that hardware is worth it! Most Linux/open source folks know better. Really, $260,000 to get lower performance than four HP blades/Win Server 2008 costing $87,000 inclusive of WebSphere licenses? Does this not alarm anyone? For other interesting comments and my responses, see: http://blogs.msdn.com/gregleak/archive/2009/05/13/latest-websphere-7-and-net-benchmark-results-stir-debate.aspx -Greg