Thanks Alex! > I have tried your test. I run your server script on a P3 700MHz machine. > "twisted" return way faster than both oldqx and newqx. I think there is a > problem with "newqx" on Opera 7.23 (on both Linux and Windows). > The gif's came > up very very slow at every other reload request. This did not happen on > "oldqx" or twisted. Interesting... Well, that wasn't what I was expecting, but I'm glad you confirm that quixote/twisted is slower than twisted by itself. That's what I'm trying to fix (And I wanted to make sure it wasn't just my machine being dumb). Sounds like I've got a way to go, though. > I didn't time them accurately (just by looking at Opera's status bar). I didn't either. I'm going purely seat-of-the-pants, here, because every time I try to slap ab on it, the problem disappears, even if I tell ab to use Keep-Alives. BTW, my perception (for the whole page with 100 images) is: twisted ~ 0.5 seconds newqx ~ 1 second oldqx ~ 10 seconds That's with the server on my 1Ghz Linux machine, and me browsing with Moz 1.7b or IE6 from a W2k box, connected by 100mbps LAN. To time them accurately, I have to use Microsoft's ACT, which is the only testing program I've come across so far that will corroborate my seat-of-the-pants feeling. I get the following results: with 2 concurrent connections (typical of browsers) twisted ~ 175 req/sec newqx ~ 127 req/sec oldqx ~ 9.9 req/sec with 10 concurrent connections twisted ~ 188 req/sec newqx ~ 135 req/sec oldqx ~ 33 req/sec I doubt many people on this list will have access to that program, which is why I put together that test package. > Btw, the locator() function return an error on my machine... here is the > traceback... Looks like my One-Size-Fits-All doesn't fit! Thanks again, Jason