[877] in testers
Re: SAVECORE and afs cache size
daemon@ATHENA.MIT.EDU (daemon@ATHENA.MIT.EDU)
Fri May 25 00:32:21 1990
Date: Fri, 25 May 90 00:31:41 -0400
To: John F Carr <jfc@MIT.EDU>
Cc: testers@MIT.EDU, Ken Raeburn <Raeburn@MIT.EDU>
In-Reply-To: John F Carr's message of Thu, 24 May 90 20:29:44 -0400 (EDT),
From: Richard Basch <probe@MIT.EDU>
Date: Thu, 24 May 90 20:29:44 -0400 (EDT)
From: John F Carr <jfc@ATHENA.MIT.EDU>
References: <9005241346.AA06265@JELLICLE.MIT.EDU>
I think it is bad to set cachesize to 1 unless you mean "flush all files
from the cache".
Indeed... now that AFS reads are designed not to overflow the cache, you
will really be crawling if you do this. Each directory will be read in,
and then flushed as the next directory along the path is read. Then
finally, that directory will be flushed and the binary will be paged in,
etc. Ezra did this once and his comment was that he wished that the
read fix was not in. However, he does agree that what happened was
indeed correct behavior; it just hurt a lot, though.
-Richard
