Experimental Physics and
Industrial Control System

"Zelazny, Michael S." <[email protected]> · Fri, 15 Nov 2002 14:40:48 -0800

Only antidotal evidence.  We've never done any measurements, but 
when compared to not doing file writes at all from the IOC, they're 
expensive.

Mike

-----Original Message-----
From: Kay-Uwe Kasemir [mailto:[email protected]]
Sent: Friday, November 15, 2002 2:32 PM
To: Zelazny, Michael S.
Subject: RE: Channel Watcher V2.0

Hello Micheal:

At 03:19 PM 11/15/2002, you wrote:
><for the IOC writing to the host,>  File writes are very expensive operations.
I agree with your statement.

For the SNS, we're beginning to use a tool called 'bumpless' 
where the IOC periodically writes setpoints to files.

Something that's running on the host, using ChannelAccess to monitor
the values but does the file writing on the host, sounds much better.
But I'm told: No, NFS is fine, and it's running at a lower priority
so it doesn't disturb the IOC.

I just think that writing files via NFS must be lame  compared to ChannelAccess:
NFS performs round-trip requests to "open", "write", "write", "write", ..
then "close".

Do you have any measurement, any proof for "file writes are expensive"?

Thanks,
-Kay

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System