Re: Meaningful comparison metrics anyone ? (@AC November 2012 17:01 GMT)
That's a load of BS right there. Mainframes have their uses, but there is no magic - as always it depends on particular use cases and there are many "buts" and "ifs".
You are gravely mistaken if you think that x86 tops out at 30% with little IO. Such a system would be wholly unsuitable for the tasks it is actually used for (in the real world) - supercomputing, rendering farms etc., usually requiring very fast and heavily used IO connectivity. Speaking of which - what do you mean by "extremely high I/O"? Hundred 8Gb FICON ports going flat-out? Or more sedate 8 or 16, which is far more usual? Because your home PC with SLI does almost the same IO as 8 8Gb FICON ports combined when you play a game. And if you are concerned with blocking IO, then you can use 160 threads on a budget with 8-socket x86.
That is not to say that x86 is on par with mainframe (it isn't), but that the comparison is not so straightforward performance-wise when smaller or midsized mainframes are concerned.
As for EAL5 - if you use zVM (which I suppose quite a lot of people do), you have EAL4+ only. The same as Windows or Red Hat Linux on x86.
One of two real reasons to use mainframe over x86 (which sadly you have completely failed to mention) is RAS. With regard to RAS, mainframe is far, far ahead of machines built on RISC/x86.
Majority of mainframe use cases however have to do with existing software, which is the second reason. Porting applications from mainframe is usually such a great undertaking that it is often seen as not worth the risk.