On Sep 14, 2011, at 11:29 AM, Anthony wrote: > On Wednesday, September 14, 2011 2:20:26 PM UTC-4, Jonathan Lundell wrote: >> >> How can the JSON times have increased (slightly) with the removal of all the >> db.py code? This seems odd. >> > > The JSON-vs-XML speed issue probably has to do with using the pure-Python > simplejson library from contrib. In class Service, we should try to import > the Python library json (2.6+), then simplejson (from the system, which might > have the C optimizations) and only then fall back on contrib.simplejson. > > This would have to be done in gluon.serializers, too, and you want to make > sure that the jsonrpclib is using the right json library. > > But why would XMLRPC be faster than JSONRPC only when Auth is disabled? If > enabling Auth adds 27ms to XMLRPC, why doesn't it add anything at all to > JSONRPC? >
I figure, chip away at the problems that have a rational explanation first, and then look at the truly weird one. Headers could be a clue. Keep in mind that the auth isn't actually doing anything, except for creating the object. What that would be faster in some cases I don't know, but it's one reason I think it'd be useful to see the request headers. Assuming that the measurement is real, and that it's really Auth() overhead, there has to be a difference that Auth could get hold of.