Argonne National Laboratory

Experimental Physics and
Industrial Control System

1994  1995  <19961997  1998  1999  2000  2001  2002  2003  2004  2005  2006  2007  2008  2009  2010  2011  2012  2013  2014  2015  2016  2017  Index 1994  1995  <19961997  1998  1999  2000  2001  2002  2003  2004  2005  2006  2007  2008  2009  2010  2011  2012  2013  2014  2015  2016  2017 
<== Date ==> <== Thread ==>

Subject: Re: frc64/frc604RT performance
From: noboru.yamamoto@kek.jp (Noboru Yamamoto)
To: tech-talk@aps.anl.gov
Cc: tatsuro@kekvax.kek.jp, katoh@kekvax.kek.jp, akiyamaa@kekvax.kek.jp, odagiri@kekvax.kek.jp, furukawa@maple.kek.jp, mkaji@maple.kek.jp, SYOSHIDA@kekvax.kek.jp
Date: Fri, 25 Oct 1996 09:11:00 +0900
Marty Kraimer wrote:
>
>Noboru Yamamoto wrote:
>>
>> Ratio of CPU performances estimated from these numbers is,
>>
>>  frc40:frc64:frc604RT=1:1.3:4.1
>
>I am not sure what should be reported but another way is % cpu
>usage which is (100 - IDLE). This would give.
>
>68040 26.8%
>68060 22.1%
>frc604 5.1%
>
>Using 68040 as base and reporting as (68040 cpu)/(xxx cpu) gives
>
>frc40:frc64:frc604RT=1:1.2:5.25

Numbers I quoted as a ratio are 

        sqrt( average((ScanPeriod CPU 040)/(ScanPeriod CPU xx))
              * average((100- IDLE CPU 040)/(100 - IDLE CPU xxx))).

Average is take over scanperiods( 1 second to 0.1 second).

>
>Not sure which way makes most sense but they give similar results.
>
>I am really suprised by the 68060 numbers. Could you please
>tell us the following:
>
>For each cpu give
>   a) Speed of main memory.
>   b) Speed and size of cache memory.
>
>Also are you sure all board support packages turned cache on?
>
>Also is it possible to make up a small test case that just
>has a loop to does an integer multiply and another loop that
>does a double multiply? Make sure compiler doesnt optimize everything
>away. This test should show raw cpu performance.
>The database test may just be dependent on main memory speed
>and size of cache.
>
I will report it later.

>
>Marty Kraimer

Johny Tang wrote:

>
>Since the EPICS benchmark includes running medm,
>100-IDEL measure takes CA clients and network
>performance facts. % CPU on scanPeriod doesn't.
>
>Noboru, can you also show the % CPU on CA tasks ?
>

At the moment, I don't have % CPU on CA task on CPU64. 
For CPU40 and CPU604, we got
                CPU40           CPU604
scan            CA event CPU    CA event
 1 sec          1.9%            0.2% 
.5 sec          3.9%            0.5%
.2 sec          9.5%            1.4%
.1 sec          19.1%           

We use 10Mb Ethernet and HP735/755 running HP-UX9.05.
Version of EPICS is 3.12.2.

>
>Johnny Tang
>
>--
>_____________________________________________
>
>Johnny Tang
>
>Tel: (757)269-7239
>Fax: (757)269-7049
>E-Mail: tang@cebaf.gov
>
>Thomas Jefferson National Accelerator Facility
>(CEBAF)
>Accelerator Software Control
>12000 Jefferson Avenue, MS 85A
>Newport News, Virginia 23606
>_____________________________________________
Noboru Yamamoto
KEKB Accelerator Control Group
KEK, JAPAN
Fax:+81-298-64-0321
Tel:+81-298-64-5309
e-mail:noboru.yamamoto@kek.jp


Navigate by Date:
Prev: CapFast follies Bret Goodrich
Next: frcX vs. mvmeX on performance Johnny Tang
Index: 1994  1995  <19961997  1998  1999  2000  2001  2002  2003  2004  2005  2006  2007  2008  2009  2010  2011  2012  2013  2014  2015  2016  2017 
Navigate by Thread:
Prev: Re: frc64/frc604RT performance KAJI Masahiro
Next: Enhanced vxWorks Variable support Andrew Johnson
Index: 1994  1995  <19961997  1998  1999  2000  2001  2002  2003  2004  2005  2006  2007  2008  2009  2010  2011  2012  2013  2014  2015  2016  2017 
ANJ, 10 Aug 2010 Valid HTML 4.01! · Home · News · About · Base · Modules · Extensions · Distributions · Download ·
· EPICS V4 · IRMIS · Talk · Bugs · Documents · Links · Licensing ·