System iNetwork

Good going Chris!

It is time to hit back. IBM has ceded the field to Microsoft! If this is a sign this free ride is over -- Yahoo!

--John

Similar to comparing a fruit fly and a refrigerator.

Microslop, oops I mean Microsoft strikes again... what I find humorous is the use of the term "ground-breaking benchmark study". What it really amounts to is statistical smoke geared to those who don't have a clue about systems. Their target is mainly those in management positions who are unfamiliar with the hardware and will be captured by the dollar signs. To them, a computer is a computer. It's like a comparing slices of bread - a hearty, satisfying chunk of wheat artisanal bread to a thin, anemic slice of white bread. They are both technically bread but are they equal in taste, substance and nourishment? I think IBM needs to return the favor and demonstrate what an anemic slice of bread Microsoft really is.

Would love to have seen one of former Penton editor John Ghrist's buyers guide product comparison charts clear up the differences and leave them with all the attendant underwear showing in the light of day.

>>>>Their target is mainly those in management positions who are unfamiliar with the hardware and will be captured by the dollar signs.



One thing to add is that these business executives and managers think that they understand IT because they use iPod, iPhone, laptop, and play computer games :-) Unlike their predecessors, the “modern” business executives and managers want to make IT investment decisions. At the very least, they want to be intimately involved in IT decisions.



A little knowledge is a dangerous thing!


Thus, it is vital that we need to expose and educate these business executives and managers on the strengths of IBM Power Systems. They need to be able to raise questions like “the number of cores, the middleware and network configurations, the scale, and even the trading applications used” when they see such advertisements.

When challenged by a kick-boxing champion, a Karate master suggested that the fight take place in a phone booth - leveling the playing field.

IBM could simply suggest that Microsoft run Websphere on an IBM BladeCenter under Linux - like the Karate master - not allowing an opponent to define one-sided terms of engagement - in this case a CPU bound benchmark - on a 16-CPU Intel Blade vs. an 8-CPU Power Server.

Microsoft's transactions per second rates puzzled me. The 8-way Power server offered 8,016 TPS while the 16-way Intel Blade offered 11,004 and 12,576 under Websphere and .Net respectively. On one hand, that seems to indicate greater CPU performance under Power, than Intel - about 1 transaction per millisecond per CPU. That sounds good. On the other hand, has anyone really ever performed a buy or sell stock transaction in 1 millisecond, in a multi-tier Web application? The reported transaction rates seem inflated to me. In my experience, there's more latency than that in just communicating with a database server on a separate platform - let alone the CPU time required to process a request in an application server. Such are the questions with benchmarks.

Is there any useful information for IBM i customers in Microsoft's benchmark publication? The $260K Power Server configuration vs. the $87K Blade configuration struck me, since both were running Websphere. That's $173K extra for the option of running AIX and other types workloads (like DB2) on the same server. With those kinds of numbers, is there any wonder that as people migrate traditional workloads to Websphere, that workloads are migrating off Power platforms?

An alternative strategy for IBM i customers would be to convert 5250 applications to RPG-based browser applications, avoid the $44K Websphere license fee (in this case), and avoid CPU & memory costs associated with J2EE runtimes. Keep everything native.

The reported transaction rates seem inflated to me.

Yes, compared to other Java Application Server benchmarks. Here's a link to a SPEC report from Oracle in March-2009.


SPEC
Benchmark


In Oracle's case, average response times ranged from .31 seconds to 2.5 seconds, and maximum response times ranged from 9.43 seconds to 13.31 seconds, which seems more real-world for Java Web applications.

Regarding Microsoft's price/performance metrics, did any one else notice that they only reported application server price/performance? They didn't even mention database server costs, even though the benchmark used two (2) 16-CPU database servers, in addition to the application servers.

Chad