On Thu, Jul 7, 2011 at 4:41 PM, an0nym <[email protected]> wrote:
>
> = instance hours vs ram hours =

I don't think this request for so-called "ram-hours" makes much sense.
 I'm willing to bet that when a GAE instance starts up on a box, a
fixed amount of RAM (the max it could use, say 128MB) is reserved for
your application no matter how much it is currently using.  Even
though an app may only use 29MB right now, it *could* expand to 128MB
on a moment's notice.  If a couple apps did this suddenly, it could
push an oversubscribed server into swap and instantly take down every
app on the box.

So even though you're only using 29MB, you are probably consuming a
full 128MB of resources.  Maybe what you really want are different
instance sizes.

> = 50k quota for datastore ops =
> That's the killer (will kill appengine :) feature of new pricing
> policy. Vendor lock-in, nosql instead of sql (I can't say it's
> worse... it's unaccustomed)... with new pricing policy you go even
> further - files instead of nosql. Yeap, that's really what datastore
> will be used like when one need to decrease number of datastore ops -
> nearly pure old file write/read. By the way, memory consumption (both
> because of bigger entities and memcache usage intensity) and cpu
> intensity will increase (maybe that's the point you made instance
> hours so expensive? predictions?).

I agree with you that the 50k quota for free apps seems small.  It's
also a bit confusing now that different operations cost different
amounts.  Greg, will this free quota be changed to a free $ amount for
operations?

My suspicion is that (at least in a threaded world) most of the cost
increase in GAE will come from datastore operations, but I am having a
hard time picturing it just yet.

Incidentally, you should already be using the datastore a lot more
like a filestore than a RDBMS.  Even ignoring costs, the latency of
your system will improve dramatically if you heavily denormalize your
data.  And because you can store treeish structured data in a single
entity, denormalization isn't nearly as ugly as it is in an RDBMS.

Basically, treat the datastore a lot more like the key/value store it really is.

Jeff

-- 
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