newbuf, nbufkv, nbufbs
d_elbracht
d_elbracht at ecngs.de
Sun Oct 14 07:42:13 PDT 2007
We have 2 machines involved with this problem.
machine1, SMP, i386, 4 GB RAM was recently upgraded from 5.4 to 6.2 cvsup'ed
2007-10-10
a partition of about 2.5 TB (gstripe -s 1048576) was newfs'ed with blocksize
of 65536 and fragsize of 8192
On 5.4, this was running for months with no problem.
On 6.2 after a few hours of high thruput (network tx and rx 400-500 Mbit
each), it became unresponsible with top showing a lot of processes with
waitmessage newbuf.
So, reset, fsck etc and it run again, only after a few hours, it became
unresponsible again, showing processes with nbufkv and nbufbs
this time, I did newfs with blocksize of 32768 and fragsize of 4096 and it's
running. Thruput is decreased to 300-400 Mbit
Note, it did NEVER show the problem on 5.4
machine2, SMP, amd64, 16 GB RAM, 6.2 cvsup'ed 2007-10-09
20 partitions involving 51 disks, all gstripe -s 1048576, newfs -b 65536 -f
8192
1 partion of 12 GB, (da3s1a) newfs -b 65536 -f 8192
after a few hours, top shows newbuf and the machine is unresponsible.
tps on da3s1a is > 500, the others are < 100
I did newfs -b 8192 -f 1024 /dev/da3s1a and it's running without the problem
(yet)
The problem seems to have to do with -b 65536 and lot's of IOPS ond 6.2
Any clue ? e.g. increase BKVASIZE to 131072 and kern.nbuf to 32768 ?
Cheers,
Dieter
More information about the freebsd-stable
mailing list