Re: FATAL: Too many queued dnsserver requests

Henrik Nordstrom writes:
> Julian Richardson wrote:
> > We have about 100 people accessing the 'net over a ropey old 64k link
> > though, which spends its life almost saturdated - do DNS requests result
> > in a lot of traffic in terms of packet size or lots of packets for a
> > single request from a client's point of view??
>
> Each DNS query consists of at least two UDP packets, one query and one
> response. If your DNS server is doing searches (not using a forwarder)
> then some queries may result is a few additional queries (one or two) to
> find the correct DNS server to ask.
...
>
> If you have a slow saturated link then long service times for DNS
> queries is normal. (long service time for any type of requests on a slow
> saturated link is normal).

  However, DNS tends to suffer in particular, because UDP doesn't get
the automatic low-level retransmits that TCP applications do - the
timeouts and retransmits will be happening at a speed dictated by the
DNS requester's own timeouts instead of at the (usually more optimized)
protocol level.

  I would guess that you are seeing a significant fraction of all
incoming or outgoing IP traffic over that link being dropped out of
your router's buffer, and then corrected by TCP retransmits. The DNS
queries being dropped won't be automatically retransmitted at the same
rate, most likely; however, you may be able to work around this problem
by finding the retry interval for your DNS server and tuning it down
until it starts performing better. Of course the best solution is more
bandwidth, but I know that's often not financially feasible (or not
politically feasible to make it financially feasible...)

  I forget if you mentioned this already: are you running a large
caching DNS server on your local network, and pointing Squid and all
clients to it? If not, you definitely should be, in this scenario. At
least that way, only that server has to be tweaked, and it will cache
all cacheable DNS data to improve your performance as much as is
feasible in this situation.

  -- Clifton

-- 
 Clifton Royston  --  LavaNet Systems Architect --  cliftonr@lava.net
        "An absolute monarch would be absolutely wise and good.  
           But no man is strong enough to have no interest.  
             Therefore the best king would be Pure Chance.  
              It is Pure Chance that rules the Universe; 
          therefore, and only therefore, life is good." - AC