> At beginning we lost about 60 seconds (between packets 93 and 111) in open > operation. The server returned 6 times NFS4ERR_GRACE on client's open request. > And there is 10 second timeout between tries. We opened mt-testbg.cgi file > there. > This file get closed in packet 122. It is strange: > > NFS4ERR_GRACE The server is in its recovery or grace period > which should match the lease period of the > server.
This sounds like something I need to get fixed at the very least. > The snoop file contains more than 2250 open operations, so the rate is about > 0.1 second per file. Do not forget, we performed about 2250 rm operations and > manu lookups too. If I am not missing anything the cadence is roughly same > across the whole snoop file. > > If we consider this "slow" we should compare it with a "fast" case... > If a 7 year old netapp can untar it in 11 seconds, I was surprised to see Solaris take 5 minutes. Perhaps that is my expectations at fault. However, I find it curious that, this morning I have gtar over nfs4 prov01:/mnt/.test9# time gtar -zxf MTOS-4.261-ja.tar.gz real 2m36.787s Lets replace gtar and nfsd, with rsync and sshd. So now everything is identical (network, servers, OS) but not using nfsd. prov01:/mnt/.test10# time rsync -are ssh /tmp/MTOS-4.261-ja 4500-07.unix:/mnt/.test10/ real 0m1.387s Yeah! That's what we want to see. Just over 1 second. If I do the same rsync to the mounted file-system: prov01:/mnt/.test12# time rsync -are ssh /tmp/MTOS-4.261-ja . real 3m44.857s What about sharesmb=on. I don't have a client with smbfs enabled, so I used my OSX desktop, which is a few networks away (all internal networks at least): lundman(/Volumes/zpool1_zero_www/.test13) time tar -zxf ~/Desktop/MTOS-4.261-ja.tar.gz real 0m24.480s Nice, smb can beat nfsd, over 100mbit link. To me that feels like the world is upside-down. Trying without ZIL as recommended by Nordin (nfs umount && zfs unmount && zfs mount && nfs mount): prov01:/mnt/.test15# time gtar -zxf /tmp/MTOS-4.261-ja.tar.gz real 0m8.453s Testing a few more times: real 0m8.284s real 0m8.264s What does that mean? Not at all what I expected, since rsync over ssh is fast, and smb is fast. Just nfsd is slow. Enabling ZIL again to see if it does as poorly as before: prov01:/mnt/.test18# time gtar -zxf /tmp/MTOS-4.261-ja.tar.gz real 2m19.512s Not entirely sure what to look at next. Lund -- Jorgen Lundman | <lundman at lundman.net> Unix Administrator | +81 (0)3 -5456-2687 ext 1017 (work) Shibuya-ku, Tokyo | +81 (0)90-5578-8500 (cell) Japan | +81 (0)3 -3375-1767 (home)
