Hi there, On Wed, Feb 25, 2009 at 5:59 PM, pedepy <paul.ro...@gmail.com> wrote: > > hey all. Although my app is far from complete, I thought Id do a > little early 'load testing' to see what it would look like. > > I had noticed awful big cpu usage times, usually anywhere between > 1100ms to 5000ms on my / page ! ... After some memcaching (I had > absolutely none before that), I was able to get it to reduce to the > mid hundreds after a few loads. Anyway, here's what i did: > > My script sent gradual requests to my app, as I understand this is how > to best test the load under gae. I dunno if these values are optimal, > they were completely random: 10 requests 2 seconds appart, 20 more > starting at 2 seconds and ending at 0.25 seconds apart, and then 15 > more 'all at once'.I fetched / and then parsed the response for all a > & img tags, and followed href's and src images. Then fetched those. I > went 2 'levels' deep like that. I did not follow any forms.
This sounds like a good way to do the load test. > My first few 'simulated site hits' ran as slow as what I had > experienced at first. But quickly enough, I saw cpu processing times > drop dramatically. Now, this brought a few questions: > > first of all, I got an 'too much contention on entity' (or something > like that) error, on my first attemps at a load test, which threw 100 > requests for everything on my site all at once. I dont know what that > error is, nor how do avoid it. Are you doing a lot of writes on a single entity in the Datastore? This can cause contention, which leads to this error. This is problematic if you're trying to scale up your application. > second, I noticed that request times read like this /'url' > 200'response code' and then a time mesure and then the cpu ms time. > What is that first time mesure ? Is it the total amount of time > spent ? Does it take into account any kind of latency on the client's > side of things or what not or does this value indicate server > ressources usage only? The logs show "[response code] [wall-clock latency] [CPU-time] [response size]". If you mouse-over the number it should show you a tool-tip that explains. The client-visible latency would be the wall-clock latency plus any network delays (which we can't measure from the server side). -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 -~----------~----~----~----~------~----~------~--~---
