> the problem appeared to be related to this, but not between the CUPS
> servers.  Specifically, since CUPS' default broadcast behavior doesn't
> work across subnets, and since most of my hosts wanting to print to
> CUPS weren't on those subnets, the client hosts had to be configured
> to poll, and that broke the cups servers.

Yeah, that configuration would cause a problem with 10K printers and
lots of clients hammering the CUPS servers. That's also not the way CUPS
is designed to work. 

The way to do this (this is now in the sysadmin manuals) for large
numbers of printers is to designate one or two hosts per subnet as
printer info servers, and have them redistribute the information they
receive to the local subnet via directed broadcast/multicast. That
reduces the load on the central servers that actually route the output
(because the definitions also include the correct URI, which won't
involve the "local" servers in any queuing). It also provides a better
failure model, as the central servers can be temporarily restarted
without risking losing printing enterprise wide. 

There's no harm to installing both CUPS server and client code on each
host. If you never define a printer on the host, the server is quiet and
does nothing, and in a pinch you can use it as a distribution point if
the usual servers are unavailable. 

You should also look at the multicast support in the current CUPS. Since
multicast traffic IS routable (broadcasts are not, at least w/o
bridging), this reduces the load on the central servers w/o the pain of
messing with distribution of /etc/printcap. 

----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

Reply via email to