[OpenAFS-devel] suggested feature...

[David Thompson]

Derek Atkins wrote:
>David Thompson <thomas@cs.wisc.edu> writes:
>
>> This is a standard cache partition size:
>> 
>> Filesystem           1k-blocks      Used Available Use% Mounted on
>> /dev/hda9               124427     89819     28184  77% /var/vice/cache

Sure:
/dev/hda9           866       881    128488+  83  Linux

>Note that if you add up the 'used' and 'available' you get 118003,
>which is an overhead (from the 124427) of about 5%.  However, I would
>not want to put a 100M cache on a partition with only 110M of
>available space.  In this case I would reduce the cache to ~95M, or
>maybe even 90M.  Note that CacheItems appears to take approximately
>4.4k per cachefile.  If you have a 100M cache, that's 100K files, or
>an additional 4.4M of storage in the cache partition of the top.

Actually, we limit the number of cache slots to 10k for performance sake.  
BTW, many thanks for the fix to allow more cache items without the massive dir 
lookup overhead  :)  I look forward to using it.

root can create ~127 Meg of data on a partition of that size.  Mere mortals 
can create ~120 Megs of data on that partition.  The file system overhead of 
10K empty files is very small (220k)

By your calculations, we should have 100Meg cache + .44M CacheItems overhead, 
in a  118 Meg partition.  Also, since the cache manager is running as root, it 
should be able to go into the 5% set-aside.  It still seems like the cache 
size specified in the cacheinfo file is being exceeded (apparently by at least 
~17 Meg or ~17%).

Dave