On Wed, Dec 28, 2005 at 10:31:56PM +0100, Srebrenko Sehic wrote:
> Try to increase kern.maxclusters. It is possible that the box crashed
> due to network buffer shortage. pedro@ commited a fix to both
> -currentand 3.8-STABLE which increases buffer on-the-fly without
> panic.
>
> Also, run a netstat -m and look at peak usage. If that number is
> reaching max, increase kern.maxclusters with sysctl.
$ netstat -m
1543 mbufs in use:
1539 mbufs allocated to data
1 mbuf allocated to packet headers
3 mbufs allocated to socket names and addresses
1538/6152/6144 mbuf clusters in use (current/peak/max)
12732 Kbytes allocated to network (27% in use)
0 requests for memory denied
0 requests for memory delayed
4896 calls to protocol drain routines
The peak to 6152 happened when the connectivity through the fw seemed to
hang. We got the following error messages in the logs:
Dec 29 16:28:54 mcu-fw-1 /bsd: WARNING: mclpool limit reached; increase
kern.maxclusters
Dec 29 16:28:54 mcu-fw-1 /bsd: bmc_io_wait fails : v=40 m=03 b=01
read_data
Increasing kern.maxclusters does alleviate the problem, but presumably
I could run into this problem again if maxclusters isn't high enough.
This build is from Oct 18 so it is likely that the aformentioned commit
isn't in my build.
-jon
