Just wanted to confirm that this issue went away, and is working great now! I did make some changes, but not any change which would seem to affect scalability. Only major change was that a big percentage of my calls was raising an exception that was logged as an error, and I fixed that.
The important thing is that it works now. There is no doubt I will be using App Engine instead of PHP running on my own servers from now on. Already I can see that the better response times are leading to a lot more visits to my MySpace app. -Bemmu On Mar 10, 4:36 am, Brett Slatkin <brett-appeng...@google.com> wrote: > Hi there, > > On Mon, Mar 9, 2009 at 5:06 PM, Bemmu <bemmu....@gmail.com> wrote: > > > First of all, thanks for the great development environment. It has > > been a pleasure to use it. > > Thanks! > > > I was testing my app when no-one else was still using it, and was > > getting response times of 0.5 - 3 seconds. My app calls MySpace over > > REST and then fetches about 100 datastore objects on each call, so > > AFAIK it is mostly waiting for MySpace to respond. > > > Then I released it to the public, and started getting 72 requests / > > second, which is what I expected because the app is visible in a lot > > of peoples' profile boxes on MySpace. What I didn't expect was that > > the response times from App Engine suddenly went to over 10 seconds, > > causing MySpace to consider it a timeout. > > Has the performance improved here? Looks like it has to me, but I want > to be sure. > > > Isn't App Engine supposed to allocate enough resources automatically > > to scale my app? I repeated the test, unreleasing and re-releasing the > > app and it clearly affects the response times. > > Part of this has to do with the way we scale out your application in > response to demand. Our system is optimized to handle natural traffic > patterns. When you go from zero to 72 requests per second in the span > of a couple minutes you are likely seeing additional latency as your > application gets distributed to multiple instances. The best way to > load test on App Engine is to slowly increase the load from zero to > some percentage of your target qps over the span of 15-20 minutes, > followed by the full brunt of the load after that. > > One of our teammates explains this idea in a talk he gave at Google I/O. > > http://sites.google.com/site/io/best-practices---building-a-productio... > > We're going to add this information to our FAQ (and maybe and article) > so people won't hit this issue in the future. Is there anything else > we could have done to make this more understandable? > > One other thing: if you realize that you need to throw even more load > at us that you already have, please read this documentation about our > quota burst limits:http://code.google.com/appengine/kb/billing.html#cpu > > By filling out the linked form we can help you accommodate more load > than our billing system automatically allows for at the present time. > > -Brett --~--~---------~--~----~------------~-------~--~----~ 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 -~----------~----~----~----~------~----~------~--~---