I can independently verify that the extremely poor (and inconsistent) performance from Backends hasn't improved. A very similar test server (collects stats in memory) on a B8 in its own Application can only handle about ~50 req/sec. I tried hand-delivering the responses to the computers but the Backend was slightly faster.
I am currently battling Enterprise Support to get a straight answer; so far I've been repeatedly told to run App Stats against a 15-line, Go-based server that performs no RPCs. Beyond that, it's generally recommended to use Pull Queues over relying on direct requests to a Backend. I'm not sure why a whole layer of indirection on top of HTTP would be *faster* than raw requests but I've experienced stranger things on AppEngine. FWIW, I haven't noticed any task latency regarding Pull Queues and they seem to have fixed the bug that prevented leasing from the same queue in parallel. Though you would still need to tolerate frequent false negatives (lease returns 0 tasks when tasks exist) and TransientErrors. Cheers, Tom Jeff Schnitzer writes: > On Wed, Aug 15, 2012 at 2:07 AM, Johan Euphrosine <pro...@google.com> wrote: >> >> I was referring to the 2000+ qps benchmark result you posted at the >> beginning of this thread, it's difficult to compare those with backend >> QPS if those are coming from running ab against localhost. >> >> It could be interesting to check what QPS you get when running ab >> against App Engine frontend urlfetching to your VPS, so we can compare >> with App Engine Backend and Compute. > > I'll try running some more tests when I get some time, but I don't see > why I would expect any different results. I might expect better > results if 'ab' isn't competing with node for CPU resources. The only > practical difference of moving 'ab' off localhost is whether the > network card driver on the VPS server is dramatically less efficient > than the loopback driver. Maybe it is, but it's likely insignificant > compared to the work done at higher levels of the network stack (like, > say, decoding http in javascript). > > This assumes that an unrestricted number of GAE frontends can pass > through any amount of load I am capable of generating, which I take > for granted. > > Jeff -- You received this message because you are subscribed to the Google Groups "Google App Engine" group. To post to this group, send email to google-appengine@googlegroups.com. To unsubscribe from this group, send email to google-appengine+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/google-appengine?hl=en.
