Experimental Physics and
Industrial Control System

Hunt Steven <[email protected]> · Mon, 16 Feb 2004 09:08:56 +0100

Benchmarks are misleading and often unreliable.  However they are also
interesting, so I have done some comparative tests on epics performance. 

I have a database of calc records processing at 10Hz, and measure the CPU
load for different IOC types.  Here are some of the results: The value of ES
is the comparative performance normalized to one of our standard VxWorks
systems.

Machine   OS      MEM(MB) CPU     Speed   L2cache Rec/sec %CPU    ES
mvme2306  vxWorks  32     PPC604                  10000   10      1.000
PC        RH73    131     MobP4   1.7GHz          100000  36      2.778
PC        RH62      ?     PIII    1.0GHz  256K    100000  74      1.351
PC        RH62    128     PII     233MHz          10000   27      0.370
PC        RH73    256     PIII    700MHz          50000   37      1.350
PC        RH73    223?    Celeron 2.4GHz  128K    50000   18      2.778
PC        RH73    256     P4      2.4GHz  512K    50000   9       5.556

I suspect that the much better performance of the Pentium 4 system compared
to the Celeron is due to having more level 2 cache rather than just the CPU.

As you would expect, the results for a particular machine do not change
(much) if you vary the number of records.

Steve

-----Original Message-----
From: Tatiana V. Salikova [mailto:[email protected]] 
Sent: Sonntag, 15. Februar 2004 17:00
To: [email protected]
Subject: How is it that IOC process only 11000 records per second at AMD
Duron 850MHz?

Hello,

How is it that IOC process only 11000 records per second at AMD Duron
850MHz?

R3.14.5 works at RedHat 8. Database consist of 40000 ai records,
synchronous devAiSoft servers all records, routine increments value
pai->val. if records is scanned with o.1sec interval, then some PVs does
not changes

"if a device causes delays of greater than 100usec
then asyn device support is appropriate." from IOC guide ch.12

Is restriction determinate RTC effect? x86 has worked with 100Hz, 68K has
60Hz?
"68040 IOC can process more then 6000 records per sec" IOC guid 3.2 basic
attributes.

RTC_x86/RTC_68K=1.7  mavPV_x86/maxPV_6=11000/6000=1.8

thank you

Best Regards,
Tatiana
------------------------------------
Budker Institute of Nuclear Physics
prospekt Lavrentev  11
630090, Novosibirsk, RUSSIA
fax     +7 3832 342163
phone   +7 3832 394977
e-mail: [email protected]

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System