EPICS Home EPICS Controls Argonne National Laboratory

Experimental Physics and
Industrial Control System

Format page
for printing


Search Tech-talk
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: Re: catime tool
From: Ralph Lange <[email protected]>
To: EPICS Tech Talk <[email protected]>
Date: Mon, 19 Mar 2012 13:35:17 +0100
For one connection between client and server, Channel Access is putting multiple requests into one TCP packet.
In your case, the packet turnaround time seems to be in the order of 32ms, and more than 100 requests fit into one packet.
So for 1000 channels and up, the requests use more than one packet, and the relation becomes non-linear until it hits saturation at a constant load of ~9us per request.

Cheers,
~Ralph


On 19.03.2012 13:31, [email protected] wrote:
While using the catime tool is have seen that the time taken per operation in any test (say channel Connect test) reduces with the number of channels used. Can anyone explain why this is so. I am attaching the datas for the channel connect test , tested with varied numbers of PVs. It reduces from 32553 us (for 1 PV) to 9.3 us (for 100000 PVs) gradually.

Testing with 1 channels named VEC:Start
Channel Connect Test
--------------------
Per Op, 32553.8880 uS (   0.0000 MHz ),   0.0157 snd Mbps,   0.0079 rcv Mbps
Search tries per chan - mean = 1.0 std dev = 0.0 min = 1.0 max = 1.0
channel name=VEC:Start, native type=5, native count=1

Testing with 10 channels named VEC:Start
Channel Connect Test
--------------------
Per Op, 3085.6591 uS (   0.0003 MHz ),   0.1659 snd Mbps,   0.0830 rcv Mbps
Search tries per chan - mean = 1.0 std dev = 0.0 min = 1.0 max = 1.0
channel name=VEC:Start, native type=5, native count=1

Testing with 100 channels named VEC:Start
Channel Connect Test
--------------------
Per Op, 345.5611 uS (   0.0029 MHz ),   1.4816 snd Mbps,   0.7408 rcv Mbps
Search tries per chan - mean = 1.0 std dev = 0.0 min = 1.0 max = 1.0
channel name=VEC:Start, native type=5, native count=1

Testing with 1000 channels named VEC:Start
Channel Connect Test
--------------------
Per Op,  41.2214 uS (   0.0243 MHz ),  12.4207 snd Mbps,   6.2104 rcv Mbps
Search tries per chan - mean = 1.0 std dev = 0.0 min = 1.0 max = 1.0
channel name=VEC:Start, native type=5, native count=1

Testing with 5000 channels named VEC:Start
Channel Connect Test
--------------------
Per Op,  19.0316 uS (   0.0525 MHz ),  26.9026 snd Mbps,  13.4513 rcv Mbps
Search tries per chan - mean = 1.0 std dev = 0.0 min = 1.0 max = 1.0
channel name=VEC:Start, native type=5, native count=1

Testing with 10000 channels named VEC:Start
Channel Connect Test
--------------------
Per Op,  12.0708 uS (   0.0828 MHz ),  42.4163 snd Mbps,  21.2082 rcv Mbps
Search tries per chan - mean = 1.0 std dev = 0.0 min = 1.0 max = 1.0
channel name=VEC:Start, native type=5, native count=1

Testing with 50000 channels named VEC:Start
Channel Connect Test
--------------------
Per Op,   9.8261 uS (   0.1018 MHz ),  52.1063 snd Mbps,  26.0532 rcv Mbps
Search tries per chan - mean = 1.0 std dev = 0.0 min = 1.0 max = 2.0
channel name=VEC:Start, native type=5, native count=1

Testing with 100000 channels named VEC:Start
Channel Connect Test
--------------------
Per Op,   9.3902 uS (   0.1065 MHz ),  54.5249 snd Mbps,  27.2624 rcv Mbps
Search tries per chan - mean = 1.0 std dev = 0.0 min = 1.0 max = 2.0
channel name=VEC:Start, native type=5, native count=1


Shantonu Sahoo

References:
catime tool ssahoo
Navigate by Date:
Prev: catime tool ssahoo
Next: RE: ARM - EPICS performance evaluation Hill, Jeffrey O
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: catime tool ssahoo
Next: BOY: Macro for workspace directory? Ralph Lange
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 
ANJ, 18 Nov 2013 Valid HTML 4.01! · Home · News · About · Base · Modules · Extensions · Distributions · Download ·
· Search · EPICS V4 · IRMIS · Talk · Bugs · Documents · Links · Licensing ·