On Jun 20, 2:13 pm, Ross Vandegrift <[email protected]> wrote:
> On Sat, Jun 19, 2010 at 02:16:42PM -0700, cd34 wrote:
> > I did a quick run of ab, several requests came in at 3.1 seconds.  I
> > don't know what sort of network they are running, but 3 seconds seems
> > very similar to an arp table rebuild in certain switches.  When enough
> > machines are put on a network, it overflows the 8192 record tcam (or
> > whatever the limit is on that switch), which causes it to need to
> > rebuild the arp tree as it rediscovers.  Until each machine is
> > rediscovered, they are basically 'off-net'.
>
> Such ARP table exhaustion would typically be steady-state in a
> datacenter environment - a complete and total meltdown, never
> recoverging.  3 seconds wouldn't be nearly enough time for such a
> thing to wring out.  Also, TCAM isn't used for ARP entries - lots of
> things are stored in TCAM, but ARP isn't one of them.

my error.  We don't run Cisco gear.  Our client modified:

sdm size ip-adjacency 8192

to

sdm size ip-adjacency 12288

which fixed their 3 second pauses on their network.  I had thought
they claimed it was a tcam setting, but, in reviewing my notes, I was
wrong.  The symptom was identical to what I was seeing when I was
fetching from his network.  I am still getting 6%-8% packet loss from
two separate locations to his current host on their inbound router.
While ICMP is a bad indicator of issues, the packet loss did coincide
with the throughput issues while downloading.

-- 
You received this message because you are subscribed to the Google Groups 
"pylons-discuss" group.
To post to this group, send email to [email protected].
To unsubscribe from this group, send email to 
[email protected].
For more options, visit this group at 
http://groups.google.com/group/pylons-discuss?hl=en.

Reply via email to