It sounds like you would be pushing any architecture to the limits.  A few
thousand requests per second with shared state is a nontrivial problem,
especially if you have a poor read/write ratio.  If you have a long-polling
scenario you may have chosen the wrong platform; this is something that
node.js or twisted python or equivalent asynchronous architectures excel
at.  But then expect to do a lot of work setting up and maintaining a fleet
of servers; it isn't a one-man job anymore.

What percentage of your (present) cost is instance hours vs datastore
operations?

> and finally I shouldn't be paying for something called experimental!

Maybe you shouldn't be *using* something called experimental.  You're
utilizing GAE in a way that is not efficient.  Possibly your application is
one that cannot be run efficiently on GAE; it sounds like an async
appserver would be more appropriate.  But maybe it can be done.  Perviously
Google hid the real cost of your app from you; now you're getting charged
based on a closer approximation of the "real cost" of service.

If you want help fitting your application to GAE's architecture then you
need to post a lot more details about what you are doing.  This post was
the first time in your countless messages that you included enough
information for any of us to even get a glimpse.  Seriously, we (the
community) are willing to help, but you need to actually want help.

Jeff

On Tue, Dec 6, 2011 at 9:08 AM, Andrius A <[email protected]> wrote:

> Did you Simon reduced the costs by 1.5-2x? I dont know what are you
> building but if I had a blog site running in GAE I wouldn't be complaining.
>
> I think I am pushing GAE to the limits, my sites do a lot of ajax requests
> with some processing on the server (1 request per user each second) and I
> have thousands of users, you would say that I should use channels? Yes, I
> thought so, but can't since there is no way to broadcast the same message
> to all users and looping through each channel will delay sending message!
> You would then say use backends to off load sending message through
> backends? Ohh yes, but backends do not support channels! So backends are
> really good, you can have them running on background and doing time
> consuming work which actually saves your money! But wait, they are not
> reliable, they restart few times per day and this is what google says in
> documentation, so you need to build some mechanism on top to make sure when
> backends restart your app do not collapse!
> So far I have moved all processing to backends and made all data to be
> passed to user requests via memcache and believe me this is where it gets
> more complicated. You later realise its nearly impossible to have backends
> and frontedns working in local environtment, so develpment process and
> debugging is really really painful. Also backends in production restart few
> times per day so reliability is very low, so on top you build  few master
> backends who look after each other and launch other backends when needed
> through task queue, and let this make more complicated, we know that task
> queue can execute in some occasions the same task few times (which is more
> frequent to happen within backends) and then you rely on memcache, and lets
> make it a bit more complicated - you know that memcache can be  flushed any
> time, so then you have datastore which can also time out and then to be it
> the most complicated to handle - request can occasionaly  fail to load. Do
> you see the point? You have a so many API's, we are being charged for them
> but they are too unreliable, thats totoally not right. I understand that
> nothing is  reliable, anything could happen, but at least backends should
> not restart few times per day, datastore shoudn't time out every hour,
> requests shouldn't fail every day! and finally I shouldn't be paying for
> something called experimental!
>
>
> On 6 December 2011 11:55, Simon Knott <[email protected]> wrote:
>
>> Andrius,
>>
>> I'm curious as to what changes you've made to your application to fit in
>> with the new pricing model?
>>
>> Whilst a lot of people said they were facing 10x price increase (some
>> were 100x) when the model was first announced, a lot of those people have
>> since come back and stated how they've managed to reduce that to 1.5-2x by
>> re-writing their code.
>>
>> Cheers,
>> Simon
>>
>> --
>> You received this message because you are subscribed to the Google Groups
>> "Google App Engine" group.
>> To view this discussion on the web visit
>> https://groups.google.com/d/msg/google-appengine/-/A7uwTwRi69wJ.
>>
>> 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.
>>
>
>  --
> 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.
>



-- 
We are the 20%

-- 
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