Re: Sun SPARCstation IPC Resurection Project!

From: Dr. David Kirkby (drkirkby_at_ntlworld.com)
Date: 09/12/03 

Date: Fri, 12 Sep 2003 06:39:32 +0100

Chris Morgan wrote:

> > http://atlc.sourceforge.net/

<snip>
> Run times: T_sequential is 110 s. T_parallel is 112 s.
> Speedup=T_sequential/T_parallel is 0.982. Efficiency=Speedup/N_cpus is 0.982
> PASS: benchmark.test
> ======================
> All 85 tests passed
> (5 tests were not run)
> ======================
>
> This was an Athlon XP 1466 Mhz cpu, Slackware 9.0

Thanks for that.

As a matter of interest, here is the same results on a Cray Y-MP
supercomputer with 4 CPUs. The execution time is not a mistake!!!
           
Hardware provider is Cray. Hardware platform is Y-MP.
Machine: CRAY_Y-MP. Sysname: sn5176. Release: 9.0.2.2. Nodename: sn5176.
Number of CPUs supported by system is unknown. Number of CPUs online is
4.
CPU_type is unknown. FPU_type is unknown. Speed of CPU(s) is 33.3 MHz.
RAM is unknown Mb.
L1 data cache is unknown; L1 instruction cache is unknown; L2 cache is
unknown
Run times: T_sequential is 14068 s. Not configured for parallel
operation.

To be fair, the code is not written to work with vector processors and I
was not the only user on the Cray.

A machine with 4 x 1.4 GHz Itanium CPUs did that in 19 s. Again I was
not the only user, but the code is multi-threaded to make best use of
SMP systems. My quad processor Sun Ultra 80 does that it 86 s using all
4 CPUs, or 265 s using just one.

If anyone can check that code on some odd hardware/software I'd be
interested. The code is not written as a benchmark, so I'm more
interested to know if it fails to compile, gives compiler warnings or
fails any self-tests.

If anyone has a dual Celeron I'd like to know if works. On one persons
dual processor Celeron, the code actually runs *slower* when running
multi-threaded and using both CPUs, than it does when using just one
CPU. I think that is a problem with the Celeron CPU - it is simply not
designed for use in SMP systems, so performs badly. 

-- 
The day Microsoft makes something that doesn't suck is probably the day
they start making vacuum cleaners." -Ernst Jan Plugge.
Dr. David Kirkby,
Senior Research Fellow,
Department of Medical Physics,
University College London,
11-20 Capper St, London, WC1E 6JA.
Website: http://www.medphys.ucl.ac.uk/~davek
Author of 'atlc' http://atlc.sourceforge.net/

Relevant Pages

  • Re: Sun SPARCstation IPC Resurection Project!
    ... Number of CPUs supported by system is unknown. ... L1 data cache is unknown kB; L1 instruction cache is unknown kB; L2 ... Hardware provider is unknown. ...
    (comp.sys.sun.hardware)
  • Re: New libc malloc patch
    ... > someone steps up to change the way mmap and brk interact within the ... > be allocated with brk. ... we already have systems running with enough CPUs that this is an issue. ... > address space and the cache: the mapping of logical pages (what you ...
    (freebsd-current)
  • Re: Purchasing the correct hardware: dual-core intel? Big cache?
    ... there's not enough IO to stress the disk subsystem. ... with more CPUs by getting true dual-core pentiums. ... The question this all pivots on is will 8M of cache be a significant ... We're looking hard at getting either Intel dual-core procs, ...
    (freebsd-questions)
  • Re: Atmel releasing FLASH AVR32 ?
    ... a cache doesn't impact other accesses to non-cacheable ... Branch prediction cost is chasing an ever eluding target. ... There are few wasted cycles on modern embedded CPUs. ... Hardware multithreading doesn't give much performance on a high ...
    (comp.arch.embedded)
  • Virtual alias cache coherency results (was: x86, ARM, PARISC, PPC, MIPS and Sparc folks please run t
    ... this test is only for one kind of cache coherency: ... CPUs or with the i-cache, are not the subject of this test). ... have a performance penalty when there is virtual aliasing. ... performance penalty for virtual aliases that aren't a multiple of 1MB ...
    (Linux-Kernel)