With Apache xmlrpc-1, each xmlrpc call corresponds with a new TCP connection -- even if its using http/1.1 and keepalive.
On Wed, Jan 14, 2009 at 11:15 AM, Mike Boyers <mboy...@yahoo.com> wrote: > I'm not sure if this helps with the discussion or not, but here goes. > > I haven't paid much attention to what happens when using httpclient within > the xmlrpc framework, but around the days of httpclient v2.0, I used it > extensively in a standalone fashion. I did look at version 3.0 as well, but > haven't looked at anything since. But I assume what applied then will still > apply now. > > Anyway, I can confirm that the httpclient framework itself supports > keep-alives. I used WireShark (which was named Ethereal at the time) and > paid pretty close attention to how it behaved. If I remember right, it has a > couple small idiosyncracies, like not being able to close it's side of a > connection at the instant that the remote side closes it, but they had > workarounds for this type of thing that prevented errors. If I remember > right, it would close it's side of the connection immediately before firing > off the next transaction. This particular behavior may be corrected now. > > I ended up using httpclient it for my project and it performed well. I used > it under pretty load, about 100k requests an hour and didn't run into any > real issues. > > I do remember that in order to enable keep-alives, the configuration was > slightly different from the most basic configuration, but it was still very > simple. > > I had hoped to have a little more time to mess around with this within the > framework of ws-xmlrpc and report results and give an example, but I just > haven't been able to make the time. > > -Mike > > > > -- http://inconnu.islug.org/~ink finger i...@inconnu.islug.org for PGP block
