[OpenAFS-devel] openafs-1.4.0-rc4 cache tuning??

Jeffrey Hutzelman jhutz@cmu.edu
Wed, 14 Sep 2005 22:24:50 -0400


You must be having an unusually bad day, Jeff.
A few corrections...

On Wednesday, September 14, 2005 08:46:37 PM -0400 Jeffrey Altman 
<jaltman@secure-endpoints.com> wrote:

>> Most of the changes except for "limiting the number of files to 5000"
>> are available on the cvs head.  (DELTA cache-autotune-20050817).

The change that is still under discussion and has not been committed is to 
limit the number of cache chunks to something in the 500000-1000000 range, 
not 5000 (partly my error).

> 20050817 is the date of the first 1.4.0 release candidate.

Perhaps, but that's not why it's included in the delta name.  I know Jeff 
knows how this works, but for others' benefit...

Every delta includes a date stamp in its name, which is usually the day the 
delta was committed (sometimes it's some other date, usually due to a 
typo).  The purpose of this date stamp is to ensure the delta names are 
unique, so that gatekeepers don't have to remember if they've used a 
particular descriptive phrase before (which can happen, if similar changes 
are made more than once).  A delta name also normally encodes what branch 
it appears on; when cache-autotune-20050817 is pulled up to the 1.4 branch, 
that will result in a new delta named STABLE14-cache-autotune-20050817.  If 
it's ever backported to the 1.2 branch, that'll be 
STABLE12-cache-autotune-20050817, and so on.

As it turns out, examining the CVS history for src/afsd/afsd.c indicates 
that the cache tuning changes have not been pulled up to 1.4.  As Jeff 
explained in his earlier message, that branch has been in "feature freeze" 
since the time the first release candidate was made available, and only 
crucial fixes are being added at this time.  I'm not a gatekeeper, but I 
suspect that after the 1.4.0 final release, we'll see quite a few things 
get pulled up that were useful but just too late to make the 1.4.0 release, 
and that those will appear in 1.4.1.


Finally... please don't use the wdelta-test program; that's an internal 
development link and in fact doesn't even exist any longer.  A better link 
to the delta in question is

http://www.openafs.org/cgi-bin/wdelta/cache-autotune-20050817