Keep it short: Who is forced to leave GAE?

[Raymond C.]

I am not asking who is not happy with the new pricing (virtually most of GAE users).

I am just asking who is FORCED to leave GAE because you cannot afford to keep running on GAE under the new pricing model.  Please (if possible) state the monthly price change as well.

And what options you are considering?

[Raymond C.]

I am one of them.  Monthly charge: $900  -> $2850 (310%)

I have been looking at EC2, every cost is clear and I can control over everything.

[Joakim Erdfelt]

We are moving 22 servers away.

Already started the process to move to AWS.

Our costs went up 2800% under the new pricing.

[Daniel Upton]

I will be forced to leave app engine. I already spend thousands per
month, but that will increase to thousands per week. I'm seeing an
increase of 300%. Unless there is a change of policy I will be leaving
asap.

[Ikai Lan (Google)]

Hey guys, just some data collection: are you guys running Python?

--

Ikai Lan 
Developer Programs Engineer, Google App Engine

plus.ikailan.com | twitter.com/ikai

--
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/-/yvS-RalUAasJ.

To post to this group, send email to google-a...@googlegroups.com.
To unsubscribe from this group, send email to google-appengi...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/google-appengine?hl=en.

[Jason Collins]

We are going from $5,400/month to $26,500/month (Python) - and this is
only one of our apps.

We are going to work hard to optimize our application because we
really like App Engine, but failing that, we may have to move
elsewhere.

j

[Ikai Lan (Google)]

Jason,

I'm thinking a lot of the biggest apparent price increases come from the fact that Python 2.5 instances are single threaded, whereas Python 2.7 with multiprocessing will serve more computing per instance. We're going to work with you to make this happen. 

The billing email queues should be working now, so I want to encourage you especially to open a ticket via that email alias. 

--

Ikai Lan 
Developer Programs Engineer, Google App Engine

plus.ikailan.com | twitter.com/ikai

[milosh zorica]

am already mostly on AWS... using GAE just on side

AWS offers the best bangs for bucks ratio so far

--
Milosh Zorica

http://www.linkedin.com/in/miloshzorica

phone: +44 20 8144 5294
e-mail: milosh...@gmail.com
skype: milosh.zorica

[Raymond C.]

I am.  Maybe check again when the new concurrency function come out.  But the new pricing is coming this month...

[Jason Collins]

Ikai,

I submitted to appengine_up...@google.com - I hope that is
the one you meant.

Yes, instance-hours is a huge increase. However, we are also seeing
large cost increases on datastore writes. Are taskqueue.add() counted
under datastore writes? We're trying to figure out why the number is
so massively high for our app (300M+ ops/day?!).

j

On Aug 31, 7:44 pm, "Ikai Lan (Google)" <ika...@google.com> wrote:
> Jason,
>
> I'm thinking a lot of the biggest apparent price increases come from the
> fact that Python 2.5 instances are single threaded, whereas Python 2.7 with
> multiprocessing will serve more computing per instance. We're going to work
> with you to make this happen.
>
> The billing email queues should be working now, so I want to encourage you
> especially to open a ticket via that email alias.
>
> --
> Ikai Lan
> Developer Programs Engineer, Google App Engine
> plus.ikailan.com | twitter.com/ikai
>

[Peter Petrov]

I've already left GAE a couple of months ago. I.e. immediately after Greg replied to me that new pricing will come into effect before Python 2.7 and multithreading. My app has short bursts of thousands QPS, and without multithreading it was clear to me that for an unknown period I'd have to pay a very high price. Today's posts here prove that I was right.

Another reason was the insanely high price for instance hours - more than 10x the industry average. Sorry Google, but your servers are not made of gold. Paying that price is simply stupid, and I'm not stupid.

I've moved to a small VPS cluster at RackSpace Cloud. I rewrote my entire app as a Node.js application (previously was GAE/Python using Kay). Very happy so far, I don't think I'll ever return to GAE.

--

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/-/MDdHgnCrDecJ.

[Richard Arrano]

I'll be leaving if some of the prices aren't tweaked, particularly the
channels. I was banking on being able to use a large amount of
channels, likely in the thousands per day. I did a double take when I
realized the new price was per hundred rather than per thousand,
particularly when channels expire after two hours and need to be re-
created. Does anyone have a good alternative to the Channel API using
Amazon's solutions?

-Richard

[Santiago Lema]

Same here. Unless Python concurrency is so magical it cancels the
instance billing effect I'll have to move everything I can move away.
Or just turn it off because the indirect platform locking is rather
efficient.

I am not in the same league as those who pay thousands of dollars per
month but rather the average small developer who sees what was a 31 $
monthly bill jump to over 500 $.

I still don't understand why Google can't come up with a pricing that
is competitive with plain old linux hosting. I appreciate the
abstraction of the DataStore to save data but if getting rid of the
hassle of Linux admin means having the constant hassle of having to
optimize for the new GAE rules then maybe I should go back in time to
2000 and get a pair of good old servers.

[Martin Ceperley]

Richard, a good alternative to the Channel API is Beacon Push (http://beaconpush.com/) we have been using it and it's dead simple and works flawlessly. It supports broadcast messaging (with channel API does not out of the box) as well as per-user messaging. Also extremely affordable, 3 million messages for $3.29.

-Martin

--
You received this message because you are subscribed to the Google Groups "Google App Engine" group.

[Srirangan]

Hi Martin, 

Thanks for sharing Beacon push. There are a couple of other such services as well. I evaluated them before opting for Channel API for Review19.

Reason being, the list of people to broadcast to is dynamic. For my app, there aren't fixed "rooms" or "groups" but the broadcast recipients changes frequently. 

Channel API doesn't impose any rules, and forces you to persist client state on the server. This initially "feels" wrong but can be a blessing in disguise as you optimize the logic to decide the broadcast recipients.

I'm not saying Beacon Push and similar services are "bad". They just didn't work as well as Channel API for my use case.

- Sri

--
Srirangan  |  About  Blog  GitHub  LinkedIn  Twitter

[Richard Arrano]

Thanks Martin, that looks like a great alternative. Do you know
anything about Amazon's SNS? Is it applicable as a Channel replacement
or am I misunderstanding it? Either way, looks like a good way to
replace the now extremely expensive Channel API. Google appears to be
pricing themselves out of the cloud computing business. And I agree
with the views of a poster in another thread who mentioned that
working with App Engine and their changing models is like trying to
hit a moving target. Developers can't and won't spend all their time
reworking their applications to avoid incurring huge charges when
Google changes pricing around.

Thanks,
Richard

[Srirangan]

>>  I did a double take when I realized the new price was per hundred

>> rather than per thousand,particularly when 

>> channels expire after two hours and need to be re-created.

Correct me if I am wrong but doesn't this charge apply only after you exceed your free quota or 8000 odd created channels per day?

[Raymond C.]

Hi Peter,

Could you share about the Database/Datastore experiences as well? What are the options on RackSpace Cloud and are they good/stable (went through the site but didnt see much info)?  I am highly interested in node.js, its just the database options concern me.

Thanks in advance.

[Raymond C.]

I have been using PubNub.  I skipped GAE Channels after knowing all the limitation, esp. its just for the browser.  PubNub has everything I need, fast and being able to message to multiple users at once.

[Raymond C.]

I dont know about Beacon Push, but with PubNub, everything could be dynamic, all "channels" are just a string which you can pre-calculate or dynamically allocate as you wish. 

[Martin Ceperley]

Richard, I haven't used Amazon SNS but from what I understand it not quite the same. Beacon Push is specifically for browser based real-time apps, using HTML5 WebSockets (and falling back on flash I believe) to push real-time JSON messages to connected users browsers. Amazon SNS seems to be more of a general backend service for powering cross platform notifications, not specifically for a browser as the client.

Like PubNub, channels in beacon push are just specified by a string and can be generated on the fly.

[Joakim Erdfelt]

Using the Java SDK.

Our "Frontend Instance Hours" is where the lions share of change is coming from.

Of the rest of the resources, only "Datastore Storage" will see a change (increase of 64% in cost)

[Prashant]

I am a student. I love GAE because it provides enough free quota to try out new ideas and run your app for free until you start making significant amount of profit.

I have an XMPP based chat app. I need atleast 100,000 XMPP stanzas/day (just to run my app without making any significant profit) which according to new pricing scheme will cost me $3 per month but the minimum monthly is $9. Now I am left with no other option but to shut my app down.

[Richard Arrano]

As far as I can tell on the new billing page, it says 100 under "Free
Quota" for "Channels Created" and then a rate of $0.01 for every 100
more channels created. I could be misinterpreting it, but it seems
clear cut.

PubNub also looks like a great alternative to Channels, I'll have to
look at the two and weigh them. On a related note, if my application
is written for webapp and Django, does anyone know if it would be a
relatively simple task to set up Django/webapp on EC2 and transition
my code? Obviously I'd have to change the database to use SimpleDB,
but again, how arduous would this task be?

-Richard

[Robert Kluin]

I've not had time to play with Python 2.7 to see how much threads help yet, but the scheduler needs work too.  I frequently see under 1 QPS / instance on low (sub 150ms) latency apps.  I may be way of the mark, but it seems like just getting that fixed would be a significant reduction in cost for us, and a better utilization of resources for Google. 

[johnP]

But Robert - you did not address the Root Question: why *should*
Google dial back the revenue knob?

On Aug 31, 10:07 pm, Robert Kluin <robert.kl...@gmail.com> wrote:
> I've not had time to play with Python 2.7 to see how much threads help yet, but the scheduler needs work too.  I frequently see under 1 QPS / instance on low (sub 150ms) latency apps.  I may be way of the mark, but it seems like just getting that fixed would be a significant reduction in cost for us, and a better utilization of resources for Google.
>
> On Aug 31, 2011, at 20:44, "Ikai Lan (Google)" <ika...@google.com> wrote:
>
>
>
>
>
>
>
> > Jason,
>
> > I'm thinking a lot of the biggest apparent price increases come from the fact that Python 2.5 instances are single threaded, whereas Python 2.7 with multiprocessing will serve more computing per instance. We're going to work with you to make this happen.
>
> > The billing email queues should be working now, so I want to encourage you especially to open a ticket via that email alias.
>
> > --
> > Ikai Lan
> > Developer Programs Engineer, Google App Engine
> > plus.ikailan.com | twitter.com/ikai
>

> > For more options, visit this group athttp://groups.google.com/group/google-appengine?hl=en.

[Srirangan]

On Thu, Sep 1, 2011 at 10:49 AM, johnP <jo...@thinkwave.com> wrote:

But Robert - you did not address the Root Question: why *should*
Google dial back the revenue knob?

The mode of operation seems to be:

1. Attract users with free / very low cost, cloud infrastructure

2. Force them to use Google specific APIs aka lock them in

3. Drastically increase prices giving users only a couple of weeks notice

4. Since they're locked in, and can't migrate their app in a couple of weeks, fleece them!

I do hope somebody from Google tells me that I am wrong! :-)

[Kaan Soral]

+1 robertk

I still have faith in Appengine, I think they should roll out
concurrency before, and make changes after that.

My daily cost increased from ~$22 to ~$55
And it is mostly cpu-hours/instance hours, I think ~0.1$ per hour is a
lot too at this stage, since python is not concurrent currently, and
rolling out the new price changes just makes things much much worse

I believe after python becomes concurrent, the costs will reduce
instead of increasing

What do Java people think about this?

On Sep 1, 8:07 am, Robert Kluin <robert.kl...@gmail.com> wrote:
> I've not had time to play with Python 2.7 to see how much threads help yet, but the scheduler needs work too.  I frequently see under 1 QPS / instance on low (sub 150ms) latency apps.  I may be way of the mark, but it seems like just getting that fixed would be a significant reduction in cost for us, and a better utilization of resources for Google.
>
> On Aug 31, 2011, at 20:44, "Ikai Lan (Google)" <ika...@google.com> wrote:
>
>
>
>
>
>
>
> > Jason,
>
> > I'm thinking a lot of the biggest apparent price increases come from the fact that Python 2.5 instances are single threaded, whereas Python 2.7 with multiprocessing will serve more computing per instance. We're going to work with you to make this happen.
>
> > The billing email queues should be working now, so I want to encourage you especially to open a ticket via that email alias.
>
> > --
> > Ikai Lan
> > Developer Programs Engineer, Google App Engine
> > plus.ikailan.com | twitter.com/ikai
>

> > For more options, visit this group athttp://groups.google.com/group/google-appengine?hl=en.

[de Witte]

Hi Ikai

I seriously hope that you did some kind of research before throwing out the new pricing model. You should be able to get these charts from the datastore by analyzing your -current- customers.

Regardless, the pricing is still extreme for thread enabled Java apps, especially if you do a lot of writes. I have an increase of 0.01 to 1.70$, this is a simple form app to register customers for downloading a package. The same can be done on any php server for 5$ a month. These kind of prices are not a big concern but if our big app (under development) will cost us 3000$ a month instead of 300$ then it is logical to move.

BTW Your email API is costing us 2000$ a month to send newsletters,... I seriously hope it is a mistake to set it to $0.01 / 100. Use a technical solution to restrict spammers, not pricing!

-Wendel

[Stefan Podkowinski]

Check out http://socket.io . A single ec2 instance should be able to serve thousands of concurrent channels easily. Especially with a asynchronous web server such as node.js (I'm using Tornado as it allowed me to port my python code easily). If you need message propagation among instances, you could use a message broker such as rabbitmq. Its probably a bit more work to setup your own solution based on that instead of using an external API but can hardly be beat when it comes to pricing. 

[Guillaume Laforge]

I've just wrote a blog post on the topic of pricing, and how it
affects one of my apps:
http://glaforge.appspot.com/article/google-app-engine-s-new-pricing-model

I had known about the new pricing model, but I was thinking my apps (a
blog and an online scripting console) with their low trafics wouldn't
really be hitting the new limits. I don't feel like paying 30-60 bucks
per month for each, that means that'd force me to move to the paid
model with $9/month (x2). That still sounds a bit expensive.

Really disappointed.

Guillaume

[Raymond C.]

Check the other posts in the thread.  Peoples using Java are seeing the similar dramatic increases as well.

And for those using Python, the preview price is having the instance charge reduced by 50% already.  

[Greg]

I'm waiting to see how multi-threading affects things. Appengine has
been a great platform so far, and I'd be very surprised if Google lets
it end up become more expensive than AWS or Azure.

Remember that with AWS (or any other VPS) you really need several
instances (in different regions) to replicate Appengine and Azure's
redundancy, and you need to manage both the servers and the systems
linking them. If you don't care about enterprise-scale scalability or
redundancy, you will certainly find cheaper options - Appengine is not
the appropriate platform for your site.

I also think it's a bit rich to talk about "bait and switch" and
"constantly changing billing". Appengine has been billed as a
technology preview all along, and now it's come into full production
they're setting realistic charges. You would have to be very naive to
assume you would get a free ride forever.

[Millisecond]

We're using Java and expect a 7x increase. With threading, and some
optimizations we should bring that down to 3 or 4x - I guess we're a
bit lucky, way less than a lot of people.

But, when you're over $1k/mo already and a small business, it's really
hard to take.

And more than the cost, the way this whole thing has been handled is
very odd. Will explore the options to go to another cloud provider,
but what a pain in the arse to move a TB of data and a handful of apps
with a collective ~50k LoC and 30 million reqs/day. Hopefully we can
find a reasonable path forward.

[Pieter Coucke]

I know that 20 cents a day is peanuts, but an increase from $ 0.2 to $ 8 a day for my app is just too high.

My app consists of bursts of messages in the queue that I want to be processed as fast as possible.  I could just set the rate to 100/s and never look back at it.  Now I need to disable my queue concurrency to avoid many instances running.

Like others in this thread, I thought the new billing wouldn't be as bad as expected.  I'm now trying java multi-threading again (hoping this issue is magically fixed: http://code.google.com/p/googleappengine/issues/detail?id=4834) and see if that helps (I still see 5 instances running now).

I use Java and moved from EC2 to App Engine so I wouldn't have to worry about peaks and adding more servers (that was before Elastic Load Balancing and Autoscaling was introduced).  Also, I didn't want to bother with OS updates but want to focus on what I'm good at (development).  I'm willing to pay more for that ease of mind.  For me the strong point of App Engine is the ease of deploying/upgrading my apps.  EC2 provides better flexibility for controlling scaling, cdn, messaging and even a scalable simpledb, but at a much higher operational cost for me.  It's easier as a developer when the App Engine team has already decided for me on which platform, database, memcache, queue, ... to use.  The learning curve was sometimes steep (transactions for entities with different parents) but I'll try to remember this as a good way of learning to create scalable sites.

The strong point of app engine (I didn't have to care about the number of servers which makes it a real cloud solution) has gone with the new pricing.  EC2 micro instances are just $ 0.02 and give me 613 MB.  I already refactored my site for App Engine so each server can die at any moment, so why wouldn't I go to EC2 (simpledb, elasticache) now?

One of my clients asked me to create a (big) site on App Engine based on my advocacy, I will probably need to rethink that decision.

Sorry for the hard work of the App Engine team, but I'm really disappointed here.

(and apologies for the not so short answer)

[Nischal Shetty]

$0.01/ 100 is the same as amazons simple email.. 

[PanosJee]

We at BugSense see a x4 increase. There is space for optimization though and we hope that after the introduction of 2.7 and some refinements to keep it at x4. But we want to release a new API and we are already planning it to implement outside GAE and just use Datastore for period flashing.

We love the 0 admin cost but the future is unsure and we have seconds thoughts about porting everything off.

[Tom Newton]

I think this might be caused by your db indexes. Do you index a lot of attributes for various entities in your application?

[Bay]

I think you might have misunderstood the 9$/month. As I understand it, it basically means that if you want more than the free quotas you have to pay _at least_ 9$/month. But it sounds as if your usage charges exceeds this... therefore you _will_ be paying the 30-60 bucks per month...

I have no idea why Google decided to make the pricing so extremely high... I personally will have to reevaluate the whole thing...