Greeting all
I'm migrating our network storage on to a FreeBSD machine with Samba 4.5, FreeBSD 11 (currently 11-STABLE built on 22nd).
I have been having some issues with Samba. Performance has been good but load, even when its doing nothing useful has been extremely high. In top I'm getting 17-18% user time and 45% system time, 0 - 1% interrupts with smbd at 100%.
I did a 5 second trace on the parent smbd process and got these results:
With that in mind I set smbd max processes to a smallish number like 40 and the load disappears but naturally clients slowly stopped being able to connect. I set deadtime to 5 and then 1 minutes to match and made no difference. I've also tried Samba 4.2, 4.3, 4.4 with the same results if not slightly worse.
The Samba logs don't seem to be particularly helpful as to determining if one or a handful of rogue clients are causing my issue. (Clients are primarily Linux machines with the kernel cifs client, mounted using mount -t cifs).
I'm out of tricks and ideas so I'm open to anything.
I'm migrating our network storage on to a FreeBSD machine with Samba 4.5, FreeBSD 11 (currently 11-STABLE built on 22nd).
I have been having some issues with Samba. Performance has been good but load, even when its doing nothing useful has been extremely high. In top I'm getting 17-18% user time and 45% system time, 0 - 1% interrupts with smbd at 100%.
I did a 5 second trace on the parent smbd process and got these results:
Code:
syscall seconds calls errors
getcontext 0.000019265 4 0
fork 2.221547074 5458 0
sendmsg 0.075918559 5461 0
setreuid 0.055482378 10922 0
setregid 0.051755381 10923 0
setgroups 0.055463542 10923 0
__sysctl 0.057374873 10923 0
getgroups 0.058816533 10924 0
getegid 0.122496351 27308 0
geteuid 0.123087113 27307 0
write 0.028582307 5423 0
wait4 0.088445110 10473 0
sysarch 0.000018445 4 0
sigreturn 0.026814389 5427 72
sigprocmask 0.080300391 16340 0
read 0.016911240 3525 0
poll 0.057053120 8983 0
close 0.027361525 5459 0
accept 0.032517063 5458 0
------------- ------- -------
3.179964659 181245 72With that in mind I set smbd max processes to a smallish number like 40 and the load disappears but naturally clients slowly stopped being able to connect. I set deadtime to 5 and then 1 minutes to match and made no difference. I've also tried Samba 4.2, 4.3, 4.4 with the same results if not slightly worse.
The Samba logs don't seem to be particularly helpful as to determining if one or a handful of rogue clients are causing my issue. (Clients are primarily Linux machines with the kernel cifs client, mounted using mount -t cifs).
I'm out of tricks and ideas so I'm open to anything.