Hi

Last night someone mentioned that their experience was that their pkg repo
server performance would go to pieces performance at around 4 or more
clients?  There seemed to be general consent and muttering about tools
in Python.

I asked on an internal forum and a few points were made:

 - property pkg/threads (see the pkg.depotd(1m) manpage) should be
   tuned to use more threads in support of higher expected client numbers

 - large installations could use the load balancing procedure at
   http://docs.oracle.com/cd/E26502_01/html/E28985/apache-config.html

 - there is also the option to use nfs-based publishers - suitable only for
   installation of clients that are close to the repo server on the network;
   in this case pkg search still goes through pkg.depotd.  This can also
   be a useful benchmark of performance without pkg.depotd in the way
   ie of your disk subsystem etc performance in retrieving all those files.

 - if the above don't help, we'd be interested in knowing more about your
   config and experience

I think all the above may have been mentioned last night.  There is one
new feature to know about in S11U2: it has an Apache-based repository
that serves all content from Apache without going through pkg.depotd.
I have no experience with it, but I'm told the benefit is essentially free
(no reconfiguration etc required).  You do need to have pkg/standalone
false to use Apache - see updated pkg.depotd and pkg.depotd-config(1m)
page included in the package/pkg/depot package.

To be honest, I always thought pkg.depotd *was* a repurposed Apache
instance!

Sorry Nathan, running it in a VBox instance probably will still suck!

Gavin
_______________________________________________
msosug mailing list
[email protected]
http://mexico.purplecow.org/m/listinfo/msosug

Reply via email to