On Jul 18, 7:01 pm, Jeff Schnitzer <[email protected]> wrote:
> This is an interesting discussion.  How many apps do you think are on
> GAE that run 100+ instances?  1000+?  While these are certainly going
> to be the "important" apps for Google, I suspect they are fairly few
> in number.  With a couple exceptions, the owners are very quiet on
> this list.

As I noted before, I have admined a 100+ instance GAE app, and know a
few companies that have or currently do run at that scale. To be
completely honest, many of those (that I know personally) have
migrated to AWS or similar cloud firms. 8 cents per F1 instance hour
is a budget breaker; AWS is much cheaper and (this is far more
important to managers) has a track record of periodically lowering
prices.

Also, you might be surprised at who reads this list. I have seen
firsthand several executives and corporate strategists read this list,
or get forwarded excerpts by their engineers. They might be "very
quiet" but they are watching (cue Jaws the movie music... )

> "Less than 1% of your customers were inconveniced. 99% of people still
> had a decent time checking out." is not very comforting.  If 1% of
> Google searches took 8+ seconds to respond, the Googleplex alarms
> would deafen people as far away as Oakland.

Just as a purely practical issue, I routinely experience long load
times for certain Google services. Search is always fast, but Gmail
can take some time to load on my iPad: when i go to gmail.com, the url
box will bounce between several different urls before loading, at
least 6-8 seconds. Is there an "alarm" form that I can fill out? :-)

> I fear that GAE is optimized for single-threaded Python2.5 apps which
> take 2s to spin up, an environment that is becoming less relevant as
> time goes on.  Given an app that takes 20+ seconds to start, would a
> high-traffic app work any better than a low-traffic app?

My experience is in Java, so I have no reference to single threaded
Python 2.5 apps. With that said, the main problem is the scheduler.

Essentially, the GAE people have to rewrite the scheduler to reflect
the following rules:
1. For apps that use 50+ instances, use the current rules for
scheduling.
2. For apps that are fewer than 50 instances, change to the following
rules:
2A: Record the startup times for the past 20 instance startups.
2B: Instances that are started up do not receive requests until the
average of the startup times + 20% has elapsed.
2C: In nonpeak hours, idle instances are free or heavily  discounted
2D: During times when GAE is experiencing high latency, the scheduler
is more aggressive in spooling up more instances. Idle instances
spooled up during this time are cheaper (30% discount seems fair...)

-- 
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