Date: Tue, 3 Feb 1998 01:15:19 -0500 (EST) From: "John W. DeBoskey" <jwd@unx.sas.com> To: freebsd-hackers@FreeBSD.ORG Cc: jwd@unx.sas.com (John W. DeBoskey) Subject: NFS V3 caching question (nfs_bio.c) Message-ID: <199802030615.AA04007@iluvatar.unx.sas.com>
next in thread | raw e-mail | index | archive | help
Hello,
I am attempting to figure out why nfs v3 caching doesn't seem to
be giving me the performance boost I'm looking for. One of the items
I've discovered is that the tail end (or lack thereof for even blksize
multiple files) of a file doesn't seem to be cached.
In a shell script, I am repeatedly cat'ing an an 8k file which
I would expect to be cached. (Yes, I am talking nfs v3 to a network
appliance F540).
cd /net/netapp/home/jwd
for i in 1 2 3 4 etc; do
cat base/h/8k.h > /dev/null
nfsstat | while read line; do
... collect & output stats ...
done
done
Between each cat, as seen above, I call 'nfsstat' and gather the
Attr & BioR Hit/Miss numbers. The output follows:
Attr Hits Misses Pct BioR Hits Misses Pct
204 43 21 12 27 225
221 46 20 13 29 223
238 49 20 14 31 221
255 52 20 15 33 220
272 55 20 16 35 218
289 58 20 17 37 217
306 61 19 18 39 216
323 64 19 19 41 215
340 67 19 20 43 215
357 70 19 21 45 214
374 73 19 22 47 213
391 76 19 23 49 213
^^ ^^
3 attr misses? 2 cache misses?
I added a couple of printf() statements to nfs_bioread() in
nfs_bio.c to help understand why I'm seeing the Misses.
nfs_bio.c: offset 0 not in cache! bp->b_blkno==0
nfs_bio.c: offset 0 was in cache! bp->b_blkno==1
nfs_bio.c: offset 8192 not in cache! bp->b_blkno==1
nfs_bio.c: offset 8192 was in cache! bp->b_blkno==1
nfs_bio.c: offset 8192 not in cache! bp->b_blkno==1
nfs_bio.c: offset 8192 was in cache! bp->b_blkno==1
nfs_bio.c: offset 0 was in cache! bp->b_blkno==0
nfs_bio.c: offset 8192 not in cache! bp->b_blkno==1
nfs_bio.c: offset 8192 was in cache! bp->b_blkno==1
nfs_bio.c: offset 8192 not in cache! bp->b_blkno==1
nfs_bio.c: offset 8192 was in cache! bp->b_blkno==1
Well, offset 0 for 8k appears to be cached correctly. I do not
understand (yet, I'm working on it) why we are even attempting to
read bp->b_blkno==1 since it doesn't exist. Yes, the file is 8k.
-rw-r--r-- 1 jwd jwd 8192 Feb 2 20:47 /net/netapp/home/jwd/8k.h
Any comments (helpful) or critiques (telling me I'm looking in the
wrong place) are welcome.
Thanks,
John
ps: If the file size is 8193, then block numbers 0 & 1 are always
cached, but the last block, 2, for 1 byte length, is never
found in the cache.
pss: I almost think I'm looking for a bad length comparison.
--
jwd@unx.sas.com (w) John W. De Boskey (919) 677-8000 x6915
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199802030615.AA04007>