On Thu, May 19, 2011 at 2:52 PM, Barry Hunter <[email protected]> wrote:
>
> With 15 minutes Google appear to be offering a compromise.

This is the problem: they haven't fixed anything they've merely
shifted the burden by compromising.

Under the current scheme, as Greg explained, full-fat Java Enterprise
apps which take 25 seconds just to initialize and therefore can't be
killed and restarted quickly take up memory which Google hadn't
accounted for. Under the new scheme where time is charged in blocks of
15 minutes the burden has been shifted from Google to writers of
efficient Python and Go apps -- Enterprise Java apps are still free
riding.

It can't possibly be in Google's best interest to have
the-next-big-thing scrappy startup subsidising Big Co's legacy TPS
reports.

How about this:

Expose a scheduler tuning knob, default off, which when enabled
reduces the kill-timeout from 15 mins to 20 seconds, and the
deadline-exceeded timeout to 20 seconds. This would be completely
unworkable for the average full-fat Java app so Enterprise customers
will not enable it, but an efficient Python app which starts in 100ms
or a compiled, statically linked Go app which starts in 20ms can be
started and stopped efficiently by the original scheduler algorithm
and be charged in units of 20 seconds. The new backends feature is now
available for anyone how needs more resources.

-- 
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