Over-running gc_thresh3

[Peter Laws]

Had to plumb up a new interface on the DHCP server (running 
dhcp-3.0.1-10.2_EL3 - I know, old and bad on several counts!) and it's now 
directly exposed to a network with /19 hungry clients.

So, we're overrunning the above-named kernel threshold, which is defaulted 
to 1024 (high-water mark for ARP table entries for those of you who, like 
me at this time yesterday, had never heard of that variable).

Any suggestions on what this should be bumped to?  It's exposed to a /19, 
so 8k+ potential clients (we're running around 7k leases) so 1024 isn't 
enough and we get this stuff:

Sep  3 13:21:24 $HOST dhcpd: send_packet: No buffer space available

... And ...

Sep  3 13:21:21 love kernel: Neighbour table overflow.



... when the water gets high enough.


I don't think it needs to go to 8192, because I've been watching the ARP 
cache and it only hits 1024 every few hours.  Would 2048 be enough?


I'm assuming I can just echo the new value to the kernel, but will it make 
anything hiccup or just up the thresholds (there are actually three 
variables that need updating)?




-- 
Peter Laws / N5UWY
National Weather Center / Network Operations Center
University of Oklahoma Information Technology
plaws at ou.edu
-----------------------------------------------------------------------
Feedback? Contact my director, Craig Cochell, craigc at ou.edu. Thank you!