Hi Peter, I currently have the "pending latency" slider set to 14.9 secs min and 15.0 > secs max. I have also set it to "auto"/"auto" per tips in other threads but > can't tell a difference. >
That's most likely your problem. Check out this video: http://www.youtube.com/watch?feature=player_embedded&v=zQ5_47zy4bY#! from 25:45 onwards. You don't want to micro-manage the scheduler too much, or it will get confused. The scheduler may not be great for low usage apps, but it can do a lot better than what you're seeing. I'd recommend only having one resident instance. We're using "1 - automatic" for instances and "auto - auto" for pending latency. We have about 1-2 requests per second during the week, but maybe only 0.2 on the weekend, so its close enough to what you have. Cheers, Per On Thursday, December 6, 2012 10:19:03 PM UTC+1, Peter Warren wrote: > > Despite having very low request volume (about 1 request / 8 seconds), low > latency (<1 sec avg) and having 3 dedicated instances, our application > still suffers from frequent warmup requests. > > See attached charts for requests, instances, and latency. > > All our app does is serve static resources and items from memcache, and I > don't see any OOM exceptions in our logs, so I don't think we're churning > instances because of memory limits. > > I figured I could try brute-forcing things and upped our dedicated # of > instances to 5. That didn't help at all and only costs us more money. Look > at the attached instances chart. Google is now using an avg of about 8 > instances to serve exactly the same volume as 5 instances handled > previously. It does look like our latency decreased a bit, but we still get > tons of warmup requests. > > I currently have the "pending latency" slider set to 14.9 secs min and > 15.0 secs max. I have also set it to "auto"/"auto" per tips in other > threads but can't tell a difference. > > So 2 questions: > 1) why does Google need more than 3 instances to serve 1 relatively > low-load request every 8 seconds??? I would think 3 dedicated instances > should be far more than sufficient for 1 request / 8 seconds. > > 2) how come the number of instances seems unrelated to load? I would think > that more dedicated servers would mean fewer dynamic instances would be > created (assuming of course that the dedicated servers are under a > manageable load). That doesn't seem to be the case. > > Thanks for any insights! > > Peter > -- You received this message because you are subscribed to the Google Groups "Google App Engine" group. To view this discussion on the web visit https://groups.google.com/d/msg/google-appengine/-/yS977aD3Z-gJ. 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.