At a high level, the IO characteristics of a pSeries and a well-configured (I say again: WELL-CONFIGURED) Intel box are very similar. The major difference is in the type of numerical applications and how the numerical applications use the processor. Intel systems do slightly better at applications with primarily scalar arithmetic; pSeries boxes excel where there are more vector-oriented operations (eg, the same operation performed on large arrays). You can erase some of that difference by sheer brute force (ie, the price premium for a pSeries will buy enough extra Intel horsepower to partially compensate), but the pSeries will be more effective for problems that benefit from large tightly-coupled SMP configurations beyond what is usually commercially available with Intel boxen. There is also a cost in algorithmic complexity to make use of less tightly-coupled MP in the Intel space.
Look at the configuration of the Blue Pacific supercomputer complex for an example: some pSeries, lots of Intel; different processing queues for different types of problems. It's a right-tool, right-job problem. Both platforms are useful; you have to use the right tool on the problem at hand. -- db David Boyes Sine Nomine Associates > -----Original Message----- > From: Linux on 390 Port [mailto:[EMAIL PROTECTED] Behalf Of > Noll, Ralph > Sent: Monday, October 13, 2003 9:02 AM > To: [EMAIL PROTECTED] > Subject: linux on p-series or Intel > > > I see on linuxvm.org > > Choosing Between Linux/390 and AIX > > what about between p-series and Intel?? > > > > Ralph >
