There is a difference between fluctuations, and running a stable
benchmark for a relatively long period of time (10min+) only to have
it suddenly start to spike tens of requests to 7+ over expected for no
reason at all. This is for requests that only only take 10-50ms
normally.

Leads me to think their "scalable" infrastructure ends up killing off
the processes even while they're serving after a while, and then it
frantically tries to respawn them, leading to load spikes that kills
off the site with high cpu quota.

On Sep 5, 2:49 pm, Wooble <[EMAIL PROTECTED]> wrote:
> Your comment on that issue that the same request should take the same
> amount of CPU the second time you run it missing the whole point of
> caching and scalability.  O(N) performance doesn't meet any definition
> of "scalable" I've seen.
>
> On Sep 5, 7:41 am, Sylvain <[EMAIL PROTECTED]> wrote:
>
> > I think the quota page is not clear enough.
>
> > And the CPU/Request warning is very hard to understand.
>
> > Even the profiler is useless and it is true that 2 same request can
> > have different results : one with a warning and the other without.
>
> > I've created this issue about CPU/Request warning
>
> >http://code.google.com/p/googleappengine/issues/detail?id=680
>
> > Regards
>
> > On 5 sep, 00:43, Thomas Johansson <[EMAIL PROTECTED]> wrote:
>
> > > The day is chunked into five pieces. Roughly 5 hours per chunk.
>
> > > The idea is that this way, even if you exhaust one 5 hour period,
> > > you'll still be able to serve the remaining 19.
>
> > > Of course, the real reason is so when they start charging, this way
> > > they'll be able to charge you for spike traffic that wouldn't
> > > otherwise exceed your daily free quota.
>
> > > Speaking of quotas, I'm really starting to wonder if there isn't
> > > something really wrong about their calculation; I did a load benchmark
> > > earlier, a simple ab2 -c 10 over a long period of time, after building
> > > up the load. Everything was working great, until suddenly, after about
> > > 40k requests, a ton of requests started spewing the dreaded cpu
> > > warnings. If I'd started a -c 100 or something then I'd understand it,
> > > but this was the exact same load pattern over a long period of time,
> > > and shouldn't result in any changes whatsoever on the google side of
> > > things. A completely stable workload, but still app engine went funky
> > > after a while and starting spewing the warnings that the requests
> > > never trigger under normal circumstances. If a completely stable
> > > workload like that will cause you to go over quota, there's no chance
> > > in hell you'll ever get even close to saturating your alotted 5 hour
> > > quotas.
>
> > > (Sorry to derail your thread a little bit.)
>
> > > On Sep 4, 11:31 pm, Sylvain <[EMAIL PROTECTED]> wrote:
>
> > > > There are two kinds of quota.
>
> > > > Documentation : Bandwidth In/out per Day  = 10 000 MB
> > > > Dashboard : Data Sent/Received = 2 000 MB
>
> > > > Could you explain the difference ?
>
> > > > Thank you.
--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups 
"Google App Engine" group.
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to [EMAIL PROTECTED]
For more options, visit this group at 
http://groups.google.com/group/google-appengine?hl=en
-~----------~----~----~----~------~----~------~--~---

Reply via email to