Unisys cranks Xeon mainframe oomph above legacy CMOS iron Mainframe maker Unisys has finally done it, pushing the performance of its Xeon-based mainframes in the ClearPath Libra family, which runs its MCP operating system, above the CMOS-based legacy Libra machines, and it won't be long before its Dorado family of OS 2200 mainframes are faster on Xeon iron than on native CMOS …
Given that the zEC12 runs at 5.5 GHz and the fastest Xeon runs 3.6 GHz and you would have an emulation layer on top of that you would not run faster.
But IBM have an emulatetor running on Linux (RDz) that you can use for Development (inhouse and as an ISV). And the engine performance is quite fine - 400+ MIPS for a 3 CP machine.
But you are not allowed to use it for production.
It could be a very fine technical solution for small mainframe installations but on the other hand - the zBC12 'only' cost ~$75,000.
I would love to be able to buy a RDz with zVM 6.x and z/OS 1.13 (2.1?) and the ability to setup and run a parallel sysplex for the fun of it......
"Given that the zEC12 runs at 5.5 GHz and the fastest Xeon runs 3.6 GHz and you would have an emulation layer on top of that you would not run faster."
Yes, because clock speed is a very accurate way of comparing performance across completely different CPU architectures.
"Which raises the interesting question: if IBM could run z series faster on a Xeon, would that be a good idea (commercially)?"
You are forgetting that IBM already transitioned their mainframe CPUs. Their mainframe CPUs are basically POWER6 cores, with a different instruction set. But they have the same logic design, execution unit, floating-point units, bus technology (GX bus) and pipeline design style as POWER6.
This trend will probably continue. New z CPUs may just be special microcode on a POWER core with lots of cache (and ECC on everything).
"Given that I doubt the mainframe register sets map very well onto the x*^ architecture this is a stunning achievement."
Amen to that, for the MCP machines they are emulating a machine where each word (I think 48bits) has a descriptor of n bits describing the current state of the word, as in - uninitialized, integer, instruction part, floating point part, characters, ... but at least the basic unit is 8bits and the arithmetic twos complement.
For the OS 2200 machines they are emulating the architecture of the 1100 series, which started life as an IBM 7094 alike, not an exact copy but similar architecture, 36bit words, one's complement arithmetic (for you non-computer people that means there is both positive and negative zero and all that entails) and given the natural word is 36bits the natural "byte" is 9bits. The emulated memory map must be a real mess.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
