dhcpd performance?

[David W. Hankins]

On Tue, May 01, 2007 at 04:36:50PM -0300, Bruce Hudson wrote:
>     Normally the bottleneck is the time it takes to ping an address to
> ensure it is not already in use. I think the timeout is one second so
> the server can handle an average of one request per second. 

I'm not sure if this is true in an older version of the server, but
ICMP doesn't 'hang' the v3 server.  We schedule a timeout event and
listen for ICMP packets, during which we can process other DHCP
packets.  The OFFER is sent on the timeout, or it is aborted if we
get a 'hit'.  However - for each OFFER pending in the timeout queue,
we make the operation to insert the next OFFER into the queue one
more node longer.

So this possibly doesn't perform well for very large numbers of
simultaneous ICMP echo's (very rare since we only do it on OFFER).

Dynamic DNS, however, does 'hang' the server, and if your DNS
server is timing out, that will be a 1 second delay.  We've
wanted to fix that for a long time.


I looked at some related stuff while answering some other questions
a long time ago, and sent a note to dhcp-users to get it in the
archive:

  http://marc.info/?l=dhcp-users&m=115334003807766&w=2

-- 
David W. Hankins	"If you don't do it right the first time,
Software Engineer		you'll just have to do it again."
Internet Systems Consortium, Inc.	-- Jack T. Hankins