Thanks Jeff - the app is 'el-cajon', version 4. Python runtime.

On Mon, Aug 31, 2009 at 3:16 PM, Jeff S (Google)<[email protected]> wrote:
>
> Hi Jeff,
>
> Ah I see, thanks for the details. I'm looking into this, would you
> mind sharing which runtime you are using, and the app ID?
>
> Cheers,
>
> Jeff
>
> On Aug 31, 1:52 pm, Jeff Enderwick <[email protected]> wrote:
>> thanks - I expected that the api calls would use parallel processing,
>> but the app/servelet itself is a single thread of execution.
>> if I have api_cpu_ms of 74, and cpu_ms of 1500, then that gives 1426ms
>> for the non-api (app/servelet usage), yes?
>> I'm trying to grok how that would happen in a single thread in 965ms
>> of wall-clock time.
>>
>> Jeff
>>
>> On Mon, Aug 31, 2009 at 11:30 AM, Jeff S (Google)<[email protected]> wrote:
>> > Hi Jeff
>>
>> > On Fri, Aug 28, 2009 at 1:08 AM, Jeff Enderwick <[email protected]>
>> > wrote:
>>
>> >> Trolling my logs, I'm coming across cases where there is extreme
>> >> (~10x) variance in cpu_ms for the exact same code flow, same GET URL
>> >> and same data (not even any intervening writes to the datastore). I am
>> >> logging my db.* function accesses, and I have factored out memcache
>> >> too. For example:
>>
>> >> 92ms, 142cpu_ms, 74api_cpu_ms, followed by:
>> >> 965ms, 1500cpu_ms, 74api_cpu_ms
>>
>> >> Q1: what could cause such a whopping delta? I am using Django, so
>> >> perhaps template compilation? I used cprofile on the SDK with a
>> >> similar data/result set, and the first page served was maybe ~2x
>> >> subsequent pages in total time. Thoughts?
>>
>> > Your idea of template compilation is along the same lines as my thinking. I
>> > can't say difinitively for this case but I would guess that you might be
>> > seeing a more expensive first request when a new instance of you app is
>> > being spun up.
>>
>> >> Q2: I am assuming the 1st number after the '200' is the wall-clock
>> >> time-to-server. As the app is single threaded ... how is it able to
>> >> burn 1500ms less 74ms in only 965ms?
>>
>> > Most API calls make calls to distributed services which parallelize work
>> > across multiple machines, so it often easy to use more CPU time than wall
>> > clock time. If you want to see where the CPU usage is coming from, you can
>> > get information about CPU quota levels at any point within your code as
>> > documented here:
>>
>> >http://code.google.com/appengine/docs/quotas.html#Monitoring_CPU_Usag...
>>
>> > Thank you,
>>
>> > Jeff
>>
>> >> Thanks!
>>
>>
> >
>

--~--~---------~--~----~------------~-------~--~----~
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