Thanks Jon for the reply. After a few days of tinkering I can see how
important it is to have that spare capacity. It's not fun to make a
request that cold starts an instance.

On Sep 7, 3:35 am, Jon McAlister <[email protected]> wrote:
> Hi Jon and Alexis,
>
> Indeed by default the scheduler will try to make space for spare
> capacity in your frontend instances. There are usually two reasons for
> this.
>
> The first is that requests do not usually come in regularly, they
> arrive in spikes and various irregular patterns. In order to
> comfortably handle the spikes without pending latency, an app needs
> more instances than would be expected by purely multiplying qps and
> latency.
>
> The second is that if the app were to receive more load, having spare
> capacity is helpful to serve the additional load without pending
> latency.
>
> The design of the scheduler is that the default automatic mode decides
> to minimize pending latency, and provide excess spare capacity, to
> help deal with the two issues explained above. Obviously, not all apps
> will want this, especially in light of the new billing formula. This
> is why there is now the ability to opt-out of this by using the
> Performance Settings options of max-idle-instances and
> min-pending-delay. With those options you can opt-out and signal to
> the scheduler that pending latency and spare capacity are not as
> important for your app as instance utilization.
>
> I hope that helps,
> Jon
>
>
>
>
>
>
>
> On Tue, Sep 6, 2011 at 1:08 AM, Alexis <[email protected]> wrote:
> > Hi,
>
> > I also have noticed something odd between QPS and latency, although my
> > numbers are not as bad as yours.
> > I generally have stats like this:
>
> > Total number of instances       Average QPS*    Average Latency*        
> > Average Memory
> > 210 total                          0.599             252.3 ms             
> > 72.1 MBytes
>
> > With a latency of 0.25sec, I'd expect to have a QPS close to 4, and
> > hence reducing the number of instances needed.
> > But QPS is still very low...
>
> > What makes QPS and Latency so different?
>
> > On 1 sep, 17:04, jon <[email protected]> wrote:
> >> Hi guys,
>
> >> I may be missing something here but looking at the below screenshot of my
> >> instances, it seems like GAE is overly eager to add instances.
>
> >> Each instance seems to be serving a request every 8 seconds approximately,
> >> which feels high. Do the numbers look right to you?
>
> >> Thanks in advance,
> >> Jonni
>
> >> <https://lh3.googleusercontent.com/-R73wUepWYAA/Tl-eYELjwMI/AAAAAAAAAA...>
>
> > --
> > 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 
> > athttp://groups.google.com/group/google-appengine?hl=en.

-- 
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