openaccess

Active Member
Jan 22, 2006
32
0
156
Hi,

We've had an issue with /home becoming unresponsive on the file system. This is on FreeBSD 6.2/6.3 machines. We've futzed with the settings in sysctl.conf that at this point its getting a bit confusing on which options we've changed and it occurred to me to just ask and see if anybody knows any 'standard' settings for boxes with 2x 2.8GHZ Xeon, 3GB RAM, SCSI RAID.

We've been using like below, and I'm wondering if we're being too aggressive on some of the maximums.

Anybody out there have a FreeBSD sysctl.conf they could share, and that seems the right 'balance' for machines roughly similar to ours?

Thanks.


kern.ipc.shm_use_phys=1
kern.ipc.maxsockbuf=1048576
kern.ipc.somaxconn=32768 # MFD0802013 for DoS on too many sockets via RST
# MFD 080213 commenting uot below to see if it helps???
kern.maxfiles=131072 # need piles for apache/mysql on cpanel
kern.maxfilesperproc=8192 # MFD080421 was 65536, need piles for apache/mysql on cpanel
# net.inet.tcp.delayed_ack=0
# net.inet.tcp.msl=7500 # reduce maximum segment life
# net.inet.ip.rtexpire=300 # expire dead sockets faster
net.inet.udp.blackhole=1 # Disable RST response on closed ports
net.inet.tcp.blackhole=2 # Disable response on closed ports
net.inet.icmp.icmplim=100 # MFD080213 Handle ICMP DoS attacks
vfs.hirunningspace=8388608 # Use RAID cache, 1/2 16MB on SRCZCR
# MFD 080213 end test comment-outs
 

nyjimbo

Well-Known Member
Jan 25, 2003
1,136
1
168
New York
I've never heard of "/home" becoming unresponsive because of these basic sysctl changes. Is this a mount point on another hard drive?. Are you sure the physical drive itself isnt becoming unresponsive ?

The only setting I see that "could" have an impact is the vfs.hirunningspace but did you change that at all ? Wouldnt that affect ALL drives and mount points and not just /home ? If you have been playing with that maybe cut it in half and see if it stabilizes the system, but I still think it would affect all partions/points and not just /home.
 
Last edited:

openaccess

Active Member
Jan 22, 2006
32
0
156
Hi, yes, we have hirunningspace customized to 1/2 the RAM available in the RAID controller. This is happening only on one box, but all the partitions are off the same RAID5 system, and we end up where we doing 'ls /home' freezes, but we can SSH in as root and look at all the other file systems.

We've dug through log files, etc, and haven't been able to see anything unusual. RAID system reports no problems, netstat/ps seem fairly normal (lots of httpd, but that is normal), etc.

In other ways, this just got us thinking about optimal sysctl.conf settings overall.

I'll try commenting out the hirunningspace next I guess.
# MFD 060630 Performance to use RAID cache, 1/2 the 32MB on SRCZCR
vfs.hirunningspace=16777216

Also, this might be more of a pure-freebsd rather than cpanel-webhosting question, but I thought disk I/O was full SMP, we're on FreeBSD 6.3, and I see from this morning reboot that the giant lock is still there?
May 4 03:03:11 cp5 kernel: iir0: <Intel Integrated RAID Controller> mem 0xfa000000-0xfbffffff irq 48 at device 8.0 on pci4
May 4 03:03:11 cp5 kernel: iir0: [GIANT-LOCKED]
May 4 03:03:11 cp5 kernel: ses0 at iir0 bus 1 target 6 lun 0
May 4 03:03:11 cp5 kernel: da0 at iir0 bus 2 target 0 lun 0
 
Last edited:

nyjimbo

Well-Known Member
Jan 25, 2003
1,136
1
168
New York
Also, this might be more of a pure-freebsd rather than cpanel-webhosting question, but I thought disk I/O was full SMP, we're on FreeBSD 6.3, and I see from this morning reboot that the giant lock is still there?
May 4 03:03:11 cp5 kernel: iir0: <Intel Integrated RAID Controller> mem 0xfa000000-0xfbffffff irq 48 at device 8.0 on pci4
May 4 03:03:11 cp5 kernel: iir0: [GIANT-LOCKED]
May 4 03:03:11 cp5 kernel: ses0 at iir0 bus 1 target 6 lun 0
May 4 03:03:11 cp5 kernel: da0 at iir0 bus 2 target 0 lun 0
I think its driver dependent. Most of giant-locked stuff has been removed but not everything. I heard that 7.0 is almost "giant-free".
 

openaccess

Active Member
Jan 22, 2006
32
0
156
iir scsi driver

well, I found this patch in FreeBSD. Unfortunately, we have to bring our 6.2 boxes up to 6.3 before we can pick up the patch. Its worth a shot anyway I guess.

Revision 1.13.2.2: download - view: text, markup, annotated - select for diffs
Fri Apr 4 16:42:17 2008 UTC (5 weeks, 3 days ago) by scottl
Branches: RELENG_6
Diff to: previous 1.13.2.1: preferred, colored; branchpoint 1.13: preferred, colored; next MAIN 1.14: preferred, colored; selected 1.19: preferred, colored
Changes since revision 1.13.2.1: +42 -26 lines

Merge rev 1.20:
Fix a buffer overflow in the processing of various SCSI commands. This
likely fixes a great number of weird problems that have been reported
with this driver.