Experimental Physics and
Industrial Control System
| 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 <2013> 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 | Index | 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 <2013> 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 |
| <== Date ==> | <== Thread ==> |
|---|
| Subject: | CPU usage of IOC too high occupied by tNet and CAS-event task |
| From: | 王林 <[email protected]> |
| To: | <[email protected]> |
| Date: | Thu, 28 Feb 2013 00:16:45 +0800 |
|
Hello,
Our LLRF control system works at 25Hz
repetition, every 40ms, 8 FPGA I/O cards respectively acquire 8 RF
cavities' runtime data, generate interrupt to CPU and transfer them to
EPICS database, data finally displayed by EPICS OPI. For each
card, runtime data consists of 8 2048-element waveforms of one cavity,
and transfered to CPU memory buffer via cPCI DMA, then to 8 EPICS waveform
records, finally displayed by EDM or CSS BOY.
Currently, we are debugging with single card (and single
cavity), and data displayed by OPI is correct, but the CPU usage is as high as
56%, monitored by vxStats1-2 module. Each I/O card have the same set of
PVs, due to 56% CPU usage of one card , I am afraid 8 cards
can hardly work together.
vxWorks spy() displays the information as follows (task with
0% usage removed),
->spy
NAME ENTRY TID PRI total % (ticks) delta % (ticks) -------- -------- ----- --- --------------- --------------- tNetTask netTask 3be4f0d0 50 11% ( 56) 11% ( 56) CAS-event 3bb998b0 180 31% ( 157) 31% ( 157) KERNEL 1% ( 7) 1% ( 7) INTERRUPT 3% ( 18) 3% ( 18) IDLE 52% ( 262) 52% ( 262) TOTAL 98% ( 500) 98% ( 500) NAME
ENTRY TID
PRI total % (ticks) delta %
(ticks)
-------- -------- ----- --- --------------- --------------- tNetTask netTask 3be4f0d0 50 11% ( 142) 11% ( 86) CAS-event 3bb998b0 180 27% ( 342) 24% ( 185) KERNEL 1% ( 23) 2% ( 16) INTERRUPT 3% ( 50) 4% ( 32) IDLE 55% ( 696) 57% ( 434) TOTAL 97% ( 1257) 98% ( 757) NAME
ENTRY TID
PRI total % (ticks) delta %
(ticks)
-------- -------- ----- --- --------------- --------------- tNetTask netTask 3be4f0d0 50 11% ( 240) 12% ( 98) CAS-event 3bb998b0 180 27% ( 563) 29% ( 221) cpuUsageTask 3bc5e4a0 255 12% ( 242) 32% ( 242) KERNEL 2% ( 49) 3% ( 26) INTERRUPT 4% ( 85) 4% ( 35) IDLE 40% ( 825) 17% ( 129) TOTAL 96% ( 2013) 97% ( 756) NAME ENTRY TID PRI total % (ticks) delta % (ticks) -------- -------- ----- --- --------------- --------------- tNetTask netTask 3be4f0d0 50 11% ( 317) 10% ( 77) CAS-event 3bb998b0 50 28% ( 792) 30% ( 229) cpuUsageTask 3bc5e4a0 255 8% ( 242) 0% ( 0) KERNEL 2% ( 78) 3% ( 29) INTERRUPT 4% ( 120) 4% ( 35) IDLE 43% ( 1208) 50% ( 383) TOTAL 96% ( 2768) 97% ( 755) NAME ENTRY TID PRI total % (ticks) delta % (ticks) -------- -------- ----- --- --------------- --------------- tNetTask netTask 3be4f0d0 50 11% ( 414) 12% ( 97) CAS-event 3bb998b0 180 29% ( 1023) 30% ( 231) cpuUsageTask 3bc5e4a0 255 9% ( 337) 12% ( 95) KERNEL 2% ( 97) 2% ( 19) INTERRUPT 4% ( 152) 4% ( 32) IDLE 42% ( 1485) 36% ( 277) TOTAL 97% ( 3524) 96% ( 756) Actually, the problem is caused by 8 waveform records, each of which has
2048 float elements and "I/O Intr" SCAN field, that is, each waveform record is
also processed at 25Hz. If I remove 7 of the waveform records and remain only 1,
the CPU usage will drop down to 28%, and if remove all, it will drop down to
3%.
The platform is vxWorks5.5.1/EPICS3.14.12.1/MPC7410(500M), and these
network related macros are redefined in my configAll.h,
#define NUM_64 250 /* no. 64 byte clusters
*/
#define NUM_128 400 #define NUM_256 50 /* no. 256 byte clusters */ #define NUM_512 50 /* no. 512 byte clusters */ #define NUM_1024 25 /* no. 1024 byte clusters */ #define NUM_2048 25 /* no. 2048 byte clusters */ #define NUM_CL_BLKS 800 #define NUM_SYS_64 1024 /* 64 byte system clusters
*/ #define NUM_SYS_128 1024 /* 128 byte system clusters */ #define NUM_SYS_256 512 /* 256 byte system clusters */ #define NUM_SYS_512 512 /* 512 byte system clusters */ #define NUM_SYS_CL_BLKS 3072 #define NUM_SYS_MBLKS 3072 With the appearance above, what do you think the problem is? Wavefrom
record should not be processed so frequently, network problem, or other
configurations (of EPICS or vxWorks) should be made?
Thanks,
Lin Wang
|
- Navigate by Date:
- Prev: Re: Installing EPICS&MEDM on UBUNTU John Hammonds
- Next: Re: caClient with PCAS Andrew Johnson
- Index: 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 <2013> 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024
- Navigate by Thread:
- Prev: Re: Installing EPICS&MEDM on UBUNTU John Hammonds
- Next: Problem building EPICS 3.14.12.3 on Cygwin Mark Rivers
- Index: 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 <2013> 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024
·
Home
·
News
·
About
·
Base
·
Modules
·
Extensions
·
Distributions
·
Download
·
· Search · EPICS V4 · IRMIS · Talk · Bugs · Documents · Links · Licensing ·