IBM benchmarketing - now wi...

Standard Trick: Comparing different classes of systems

The standard marketing tactic is: When you want to make a statement on price use the cheapest system, when you want to make a statement about performance compare with the most expensive high-end system.

This is a nice example from an actual IBM marketing campaign.

This table is from the IBM website. They want to suggest that the p570 and the T5440 are similar systems. But they compare the performance of several systems in this table. In one of the following sentence they write:

Compare the 32-core 4.2GHz Power 570 to the 32-core T5440 and again the winner is clearly the 570 with 1.9 to 2.7 times the performance."

They only forget to mention, that the 32 core 4.2 Ghz Power 570 is a 16 RU system, while a T5440 takes just 4 RU. A system comparable to the T5440 would be the p550. It´s the in the four socket class, too. And what do you see. Half the SAP SD perfomance and half the SPECjbb performance.

On the top end they use the p570 to state a huge number for performance and to tell you some stuff about RAS. Of course the p570 is better here. It´s a much larger system with different design goals. But they forget to state an additonal fact in their marketing stunt: The price of the p570. They didn´t state the price of the p570 throughout the document. Just to give you a hint: Look at the TPC-C benchmarks for p570 systems. The activation of memory or one or two additional procs may cost as much as a complete T5220 or an T5440 in a smaller configuration. BTW: When you buy memory (the hardware) for a p570 you cant use it. You have to buy a memory activation as well.

So make it short: Such a comparision is just outright nonsense.

A really strange pricing comparision

By the way: The comparative pricing is really strange. The IBM states in the artcle, that an T5440 costs you$235,705. But when you look at Sun.com you will find a different pricing: An T5440 costs you $132,995.00 list price with 128 GB and 4 8-core procs at 1.4 GHz (that´s almost full blown). The solution of this miracle can be found in the Footnote 7.

At first they state:

Prices based on configurations used in the SPECint_rate2006 benchmark including processors, memory and disk.

Almost at the end of the footnote, they state:

Sun SPARC Enterprise used Sun StorEdge 6140 Array for disk.

Okay, that´s fine with me, albeit disk configuration is irrelevant for SPECint. I assume it was just ease of configuration for the benchmarking guys. But hey, it´s an Oracle pricing comparision and a 6140 makes sense. So this pricing for the T5440 was increased by the price for the 6140. Additional they took the the 256 GB configuration.

But - and now it gets interesting - they state for their own system:

All IBM systems used internal storage.

The disk subsystem for pSeries 560: 4x146 GB SAS 15K RPM. Memory? 64 GB. Nice try ... really nice try.

The system isn´t comparable at all. SPECint configurations aren´t any hint for a Oracle configuration. Thus is pricing comparisions about the system configuration are totally nonsense at all.

I can promise you one thing: At any realistic database size the T5440 configured for this comparision system will outperform the p560 configured for this comparision by a vast margin. There is one sizing rule in database sizing different to SPECint. You can substitute memory just by more memory. And two boot disks aren´t really comparable to a 12-disk shelf with redundant FC controllers and 2 GB of NWRAM cache. At any working set size beyond 64 GB the p560 would wait with all its gigahertz for the hard disk. And even with a workingset size smaller than 64 GB the 6140 would outperform the 4 disk configuration at write loads due to it´s NVRAM cache

Software pricing - or: The Oracle Problem

It´s correct, that Oracle is one of the weak points of the CMT technology. Or to be exact - not the Oracle software is the problem - Oracles licensing model is in favour of a few big cores. IBM doesn´t get bored to point on this issue. But they forget some important facts. At first: There is other software than just Oracle, and Oracle is one of the last strongholds of per-core licensing. You will see "per-seat", "per-employee" or "per-named user" everywhere. SAP? Per user! The Oracle "Application Specific Full Use" licenses for SAP? Per user! Lotus Notes or Lotus Domino? You can license it per user! Several other software products? Per User!

So it´s not relevant how many cores you have, it´s relevant how many people use a software at your company. And this enables you to use the most efficient technology over the long time.

And they forgot to mention another alternative at all. The T5440 is eligible for the Oracle Standard Edition. The advantage: A per-socket licensing. The p560 16-core is roughly comparable with the T5440 in SAP-SD .... but it´s a 8 socket system, thus you have to use the Enterprise Edition.

You think this is a non-argument? Maybe ... but i saw an IBM presentation using the same argument to sell their QCM modules by using Standard Edition on their systems and Enterprise Edition on ours. We can play the same game.

You should think about an additional factor: Oracle licenses are perpetual. Let´s assume you have an older system in your data system you want to migrate to a newer one (For example your older SAP system on a 6800). You have already Oracle licenses for some amount of cores in your house. You can transfer this old licenses to a new system.

Additionally we should consider here, that we compare list prices. Oracle sales reps are not nescessarly friends of IBM because of the DB2 menace in their account. You could use that.

And at the end: Many applications can run on Mysql or Postgres as well. There are licensing models for Mysql allowing you to install as many systems with mysql as you want at the price of a single Oracle core license. When you really want to save money, you should think in this direction.

The 11% claim

IBM likes to claim that they had an 11% increase of pSeries business. What they forget to write: They have merged iSeries an pSeries business to a new pSeries spanning both of the two old product lines. The iSeries lost 82%. Well ... i leave it to you where this increase of 11% may find its explanation.

IBMs AIX vs. Solaris claims

Well just a few notes: The point that Solaris implemented RBAC in 2000 isn´t an excuse for implementing it in AIX just this year. The WPAR Live migration is just a hack, as i explained in this article and Volume Managemnt doesn´t entered Solaris with ZFS, we had an product called Solstice Disk Suite dating back into the last century There are many "

Conclusion

I will stop to dissect the IBM statements here. There is more utter bullshit in this website from IBM but read it for you self. You will find it at their website.