That is indeed an interesting suggestion, but I suspect it might be
unworkable at that granularity. If you have an idle instance occupying
20 MB of a 128 MB instance class, GAE can't make use of the remaining
108 MB without restrictions. If your instance suddenly needs more than
20 MB to process an incoming request then it could not have it
available (without paging) because some other process would be using
it.

Perhaps a clever combination of scheduling and non-guaranteed QoS
instances would allow such finer granularity. A statistical analysis
could allow various apps with intermittent usage spikes to be combined
on the same servers / instance class partitions, in such a way that
you would known with a given probability that no trashing (massive
paging) would occur. Different probabilistic QoS / performance classes
could be worked out.

BTW, some QoS questions:

1) regarding backend classes: I suspect the B8 4.8 GHz is a 2.4 + 2.4
GHz dual core. Is it so? shouldn't that be advertised? It's not the
same for some apps.

2) What exactly is, anyway, the provided QoS of the various instances?
A B1 with 600 Mz is surely a server being partitioned in various class
configurations. But Linux is not a real-time OS, so the other
processes probably can have a significant worst-case impact on your
apps' performance. What that impact can be should probably be better
disclosed.

-- 
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