On 2/27/12 8:16 PM, "Greg Troxel" <gdt%ir.bbn.com@localhost> wrote:
...
>
> I would suggest doing ktrace.
>
> It may be that there is a raw socket (proto ospf is not UDP) and a
> packet is sent on each interface in quick succession and the raw socket
> needs bigger sendspace. I don't know if udp sendspace is reused for raw
> sockets.
Indeed, changing the UDP sendspace didn't have any tangible effect. Once
traffic picked up this morning it started throwing the error with about the
same frequency as before.
net.inet.udp.sendspace = 18432
net.inet.udp.recvspace = 83200
There are eight interfaces on the system but only two are participating in
OSPF. There are about 1000 routes being announced on the troubled interface
with OSPF. The systems was running quagga-0.99.15 but then updated to
0.99.16 as I had the package handy. I could take it to 0.99.20...? I did
update the kernel from 5.0 to 5.1.1_PATCH with bigger INQ but that also
didn't seem to change.
Attached is a ktrace/dump of an event. The time rolling up to is includes
the select() and a read off the network prior to it trying to write.
Your thoughts are appreciated,
peter
Attachment:
ospfd.txt
Description: ospfd.txt