slow connects to sv.cache.nlanr.net

Over the past few days, I've noticed that there are frequent periods
where connects to sv.cache.nlanr.net take a long time to complete,
sometimes even hitting squid's connect timeout (currently left at the
default 120 seconds). I am using v2.2 STABLE5 under OpenBSD with
query_icmp on and sv.cache.nlanr.net as my closest parent (both by
hops and ping round trip time). I have only recently started using
sv.cache.nlanr.net as a parent, so I don't know whether it's been
having problems for a while.

During these periods, the pattern seems to be as follows:
- Initial requests to a site generally goes through sv.cache.nlanr.net
  because it's the closest parent.
- Requests are observed to take a long time to complete.
- netstat shows sockets to sv.cache.nlanr.net are often waiting a long
  time for sv to accept the connection (SYN_SENT).
- Respones to ICMP queries, however, come back from sv very promptly.
- Once a connection actually times out, requests get sent direct
  for a while (they don't appear to get sent to my other parent,
  sd.cache.nlanr.net, instead).
- Requests that are sent directly or to my other parent are processed
  in timely fashion.

So, does anyone know what's up with sv.cache.nlanr.net? And is
there anything I can do to minimize the delays for my users when
one parent goes through periods where it takes a long time time to
accept connections (but remains up and responsive to ICMP queries)?
(Reducing the connect_timeout results in a lot of direct accesses,
with my other parent remaining seldom used.)

Thanks,
-ccwf
Received on Thu Jan 13 2000 - 11:22:50 MST