CMOS Processor Benchmarks

[Graham Gold]

Suspect I know the answer, but, does anyone have any benchmark information for the CMOS processors used in the ClearPath range up until the Libra/Dorado 800 series systems?

CPU benchmark or a workload/OLTP benchmark would be ideal - I had a look at the usual suspects, (TPC, SPEC) and the only Unisys results ever posted were for the Unisys windows server products (SMP, ES etc) and more recently, Forward! running SuSe Linux.

Regards,
Graham

[Paul Kimpel]

---------- Original Message ----------

*Subject:* CMOS Processor Benchmarks

*Sent:* Friday, 20 May 2016, 02:09 PST-0700

*From:* Graham Gold <thedo...@gmail.com>

The only recent performance statistic I have is that the ClearPath Libra
880/890 produced 600 Gartner MIPS for a single CPM and 5800 MIPS for a
16-CPM system. That is not a pure processor measure, and I am sure is
derived from a series of mixed-load benchmarks, so it's probably in line
with an OLTP benchmark.

This came from a 2013 UNITE presentation by Andy Beale and Steve Koss.

Paul

[Graham Gold]

Thanks Paul,
I'll look into Gartner information around MIPS then.

In the meantime, does it sound reasonable to equate 1 MIPS to 16,666 tpmC (transactions per minute on the TPC-C OLTP benchmark) - my simple maths of dividing MIPS by 60 to get a per minute value?

By that measure, a 2 module Libra 890 with 8 processors and total of 3,750 MIPS would benchmark at around 6,250,000 tpmC, unrestricted by the governor.

That would put it 5th in the TPC-C benchmark if performance was the only consideration. Does that feel about right?

http://www.tpc.org/tpcc/results/tpcc_results.asp?print=false&orderby=tpm&sortby=desc

Regards,
Graham

[Graham Gold]

Actually my maths was off, it would equate to 62,500,500 tpmC and top the TPC-C benchmark by over 30,000,000 tpmC - maybe it's not reasonable to equate 1 MIPS to 16,666 tpmC after all - I may be comparing apples and pears!

Regards,
Graham

[Paul Kimpel]

I can't speak to your comparison quantitatively, but qualitatively an
8-CPM 890 is one damn powerful system. There are other configuration
components that would affect the results, too, of course, including the
I/O channel and disk configuration, the amount of memory, and how much
of that memory was devoted to disk cache.

The other thing to consider is that the MIPS numbers Unisys reports are
a total-system performance measure, not just a processor measure. In a
benchmark like TPC, I/O performance is probably as important as CPM
performance, and if there's one thing Unisys knows how to do well, it's I/O.

Still, a system like that topping TPC-C by almost 2x seems unrealistic.
We'll probably never know for sure.
--
Paul

[Graham Gold]

I had a look at the TPC-C spec - a new-order transaction which forms the basis for the tpmC rating would encompass several instructions and the number of instructions would depend upon the OS etc so my simplistic maths was never going to be correct - as you say, will never know unless Unisys start taking part in the benchmark for ClearPath Libra/Dorado.

Thanks for your help though!

Regards,
Graham

[mike...@fisglobal.com]

The MIPS/RPM rating uses multiple variables to calculate. This includes RPM factor (24.3), # of CPU's on your host, RPM rating (IK IPSHOW ALL) and the number of days in the month...