thanks for info Nick, I realize I'm probably looking too deeply into your explanation above. I thought you had said it was not instance startup cost (which had me concerned), although I know it can be difficult to debug/profile what is actually happening.
of more concern to me is marzia's comment that 100-300ms is now considered *intensive*. this is not how it was portrayed in the past so it seems to me that while they added the ability to have longer running requests, they've also reduced what you can accomplish for the average request. I have yet to see these governors kick in but I know that a lot of my average requests fall into the 100-300ms range so I wonder how it will affect performance when a large amount of concurrent load is added. maybe it wont, your comment about piping everything through a single handler (AMF gateway) may, as you said, be part of the problem. still, I'll go back to the log warnings from the past 6 months, they just did not appear for requests under 300ms. and now I'm not sure if I'll be penalized for them btw, I have no problem with having to pay more for this, I just dont know if this portion of the CPU pricing is extensible or not... cheers brian On Feb 25, 9:47 pm, Nick Winter <livel...@gmail.com> wrote: > Thanks for following through on this, Marzia! That makes a lot more > sense. I really appreciate the diagnosis; you rock! > > bFlood, I think the only reason that handler is getting mauled is > because of the 1200ms+ startup costs for initializing a new instance. > Because it's the same handler, the startup costs are causing it to > look like a dog and get collared, delaying also the 30-200ms type > requests. > > Also note that there are lots of what are "different" requests to me, > in that they do much different things, but they share a common handler > and URL (in this case, a PyAMF gateway on /gateway). So that's why > many different parts of the site seemed slow together. > > Now I just have to find a way to hack down that PyAMF startup time. > Ugh. > > On Feb 23, 2:54 pm, Marzia Niccolai <ma...@google.com> wrote: > > > Hi, > > > Upon some further investigation, it seems that this is the result of the new > > handling of CPU intensive requests, more information about which can be > > found here:http://code.google.com/appengine/docs/quotas.html#Request_Limits > > > Specifically "Applications that are heavily cpu-bound, on the other hand, > > may incur some additional latency in long-running requests in order to make > > room for other apps sharing the same servers. " > > > Essentially, if we observe that you have some heavily cpu-bound requests, > > your handler may experience additional latency. This may not always happen, > > and for the higher cpu request handlers, there is no way to know exactly > > when it may happen. > > > -Marzia --~--~---------~--~----~------------~-------~--~----~ 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 -~----------~----~----~----~------~----~------~--~---
