[eepro100] KERNEL: assertion (flags&MSG_PEEK) failed at tcp.c(1463)

Donald Becker becker@scyld.com
Sat Apr 27 00:34:01 2002


On Sat, 27 Apr 2002, Wei Keong wrote:

> I've another server down with the same rx buffer problem again. The funny
> thing is both servers were down in the morning (5-6 am) where processing
> is the lowest.

Do you have any cron jobs scheduled for then?

Check /etc/cron.daily and /etc/cron.weekly

> Don't really understand why the kernel has not enough
> memory for the network card...
> Will tuning VM solve the problem? What are the possible side effects? I
> still have a couple of server (with similar configs) and I'm not sure if I
> should tune the VM, as a preventive messure. Please advise.

The kernel's VM subsystem can run short of available memory for kernel
needs even when the machine has lots of memory.  Paradoxically, lots of
memory seems to lead to worse short-term predictions of memory needs.
Or perhaps it's a reduced ability to scan the page list for free pages
that can be used for network buffers.

[[ Mandatory gripe: We didn't have these problems with the VM system in
the pre-2.2 kernels. ]]

Anyway, tuning the VM parameters can definitely improve the situation.
There should be a larger reserve of free pages when your machine has
lots (64+MB) of installed memory.  Having the kernel keeep 1-5MB ready
for instantaneous use is very reasonable.

-- 
Donald Becker				becker@scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Second Generation Beowulf Clusters
Annapolis MD 21403			410-990-9993