Isn't .08/hr 1.92/day $59.52/month for a 600 MHZ CPU instance with 128 MB memory a LITTLE EXPENSIVE

451 views
Skip to first unread message

John

unread,
Dec 14, 2011, 7:50:06 AM12/14/11
to google-a...@googlegroups.com
Just started thinking about this... but now that we are basically paying for all the datastore reads/writes, and bandwidth separately...

Isn't paying $60 a month for a 600 MHZ instance with only 128 MB ram a little expensive?

Just taking a quick glance at EBAY, I can buy a > 2 GHZ machines with over a GB of memory all day long.
I can buy BRAND NEW Intel Atom Dual-Core D525 Processor(1.8GHz, 1MB L2 Cache), Support Intel Hyper-Threading technology,
with 1GB memory for ~ $160 all day long

Call me crazy, but I still have my 1 GHZ pc I bought back in 1999 (12 years ago) sitting in the garage and I would have a problem giving it away (It also has a lot more memory than 128 MB ram).

A standard (small) SAME PRICEd Amazon EC2 instance comes with 1.7 GB of memory and even their FREE micro instance gives you 613 MB of memory.

I understand computers were a lot more expensive back in 1999, but they have gotten a lot cheaper over the past few years.

Please justify what I am paying for because right now I am trying to justify upgrading to the F2 instance class for twice the price ($120/month) just so I can double up and get a whopping 256MB ram!

Joshua Smith

unread,
Dec 14, 2011, 9:30:47 AM12/14/11
to google-a...@googlegroups.com
You are paying for the most excellent googly software and most excellent googly maintenance personnel and most excellent completely free support and vibrant community on this newsgroup.

If you don't think those are worth the coin, then go use EC2. Don't let the door hit you on the way out.

I've frankly had enough of the whining about "expensive computers."

Think of it this way: google is giving your all this computer time for free! All you are paying for is the license to use their software, which they bill based on usage. Just like most commercial software (Oracle, for example, charges you based on the # of "cores" you deploy).

Since the computer time is free, it is cheaper than EC2, Rackspace, and your stinkin' garage. The software is expensive, but that is completely justified.

QED.

-Joshua

--
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/-/U8nQaYwxQr8J.
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.

Rishi Arora

unread,
Dec 14, 2011, 9:50:29 AM12/14/11
to google-a...@googlegroups.com
That's one way of looking at it.  But, if you configure your app properly, and design your system properly, you pay $60 a month for that 600Mhz instance only if you actually use it at 100% utilization 24 hours a day, for the entire month.  Here's how I can illustrate that point: My app has a Max-Idle-Instance=1, but the actual number of active instances is anywhere between 5 and 30 at any given time. And it responds to changes in demand - number of requests that  have to be served. So, I'm paying for only one instance, but I'm getting the right to spawn many more.  I could easily set my Max-Idle-Instances to 30 in anticipation for worst case load, but most of those 30 instances will be idle most of the time, and yet I'd be paying $60/mo for each of them.  In other words, $60/mo buys me the right to use that instance whenever I want, at a moment's notice (no initialization latency).  I think that's justified, given that it gives you uptime guarantees, redundancy, scalability, and you can work around it so easily (by making conscious latency-vs-cost trade-offs).  Basically, Google is incentivizing you to optimize your app so that you have instances running only if and when you need to.

mike hershey

unread,
Dec 14, 2011, 10:28:02 AM12/14/11
to google-a...@googlegroups.com
I'm with you, this absolutely nuts. I got on board before all this new pricing stuff, round 1 new pricing just barely didn't scare me away, but this definitely did. I'm getting off app engine ASAP. So much about app engine seems like a scam. $1/million database writes, BUT a write is AT LEAST 2 writes when they charge you. Why can't we call it what it is? Make it $5/million writes and all writes are 1 operation. I know then your not rewarding people who remove composite indexes and optimize but it just seems dishonest that there are no write operations that use just 1 write. 

Also in my app app engine often spins up idle instances (that I cannot get rid of no matter what I configure) and send them exactly 1 request every 15 minutes so that I'm being charged the whole time for this instance I don't want.

I hung around hoping things would get better, but I'm off now.

mike hershey

unread,
Dec 14, 2011, 10:34:56 AM12/14/11
to google-a...@googlegroups.com
Also justifying the price with the features of app engine is a horrible justification. App engine is a cloud; the whole idea of cloud computing is to save money by reducing idle server time. Every other PaaS does the same thing a lot cheaper. Lets look at Amazon's elastic beanstalk: 
Service and ResourceUnitCost BreakoutCost
Amazon EC2 t1.micro instance1$0.02/hr * 24 hours * 30 days$14.40
Elastic Load Balancer1$0.025/hr * 24 hours * 30 days$18.00
Elastic Load Balancer Data Processing15GB$0.008/GB * 15GB$ 0.12
Elastic Block Store volume8GB$0.10/GB * 8GB$ 0.80
S3 Storage for WAR File and Access1GB$0.14/1GB + $0.01 for<1k PUTs, <10k GETs$ 0.15
Bandwidth In and Out15GBInbound is free, 15 GB out * $0.12$ 1.80
  Total Monthly Cost without Free Tier$35.27
  Total Monthly Cost with Free Tier$0
Way more free, way cheaper. Don't get me wrong I'm only ranting because I really want to be able to stick with app engine, but When its orders of magnitude more expensive, I just can't justify it :( What happened to not being evil?

Jeff Schnitzer

unread,
Dec 14, 2011, 10:36:41 AM12/14/11
to google-a...@googlegroups.com
I think there is a legitimate gripe here which is that large-memory
instances are unreasonably expensive.

There's some significant value-add for GAE's "whole package" -
automatic scaling, memcache, edge caching, deployment system, API
access (although these APIs are generally charged separately). This
makes the $60/mo for a basic (multithreaded) instance worthwhile.
It's expensive but it's convenient, and most frontend work fits fine
in the F1. Also it's a little bit of apples/oranges because the GAE #
is heap whereas an Amazon # is VM size, but this is probably less than
a factor of 2 difference.

On the other hand, there are many application components whose primary
requirement is a significant chunk of RAM. All that Google
infrastructure is nice but it isn't nice enough to warrant a 10X
premium just for a measly 1G of RAM. And you can't even get more.
Seriously, a cheap amazon "standard" instance has significantly more
RAM than the most expensive GAE instance... lame.

Consequently, backends are useful as a long-running frontend, but
absolutely useless as an in-memory index. We're priced into going the
inconvenient route of placing memory indexes in other cloud services.

I've been generally accepting of GAE's recent pricing changes, but the
price of large-memory instances basically means I have to treat that
option as if it doesn't exist. Which means when Google adds all these
fancy features to support different kinds of instances, from my
perspective, they're wasting their time. I can't use them until they
make them cheaper.

So here's my plea: a 256MB instance shouldn't cost twice as much as a
128MB instance, and a 512MB instance shouldn't cost twice as much as a
256MB instance. The price curve should drop off. There's a
reasonable premium to pay for running on GAE, but a factor of 10 isn't
it.

Just for comparison... the largest GAE backend, at 1G, costs $460/mo.
A 1.5G linode instance costs $60/mo. And I can get a 4G linode
instance for $160/mo. And while it's not exactly an apples/apples
comparison, when I need RAM, the priority of all those other Google
niceties goes down considerably. And if I needed (say) four 1G
backends, you can absolutely bet that I will go with Linode and pocket
the extra $20k per year.

Jeff

JH

unread,
Dec 14, 2011, 10:54:51 AM12/14/11
to Google App Engine
I've seen it mentioned here before that Google's RAM is made with
solid gold. Not sure if it's true or not...

On Dec 14, 6:50 am, John <sc...@peoplepedia.org> wrote:
> Just started thinking about this... but now that we are basically paying
> for all the datastore reads/writes, and bandwidth separately...
>
> Isn't paying $60 a month for a 600 MHZ instance with only 128 MB ram a
> little expensive?
>
> Just taking a quick glance at EBAY, I can buy a > 2 GHZ machines with over
> a GB of memory all day long.
> I can buy BRAND NEW Intel Atom Dual-Core D525 Processor(1.8GHz, 1MB L2
> Cache), Support Intel Hyper-Threading technology,

> with 1GB memory for ~ $160 all day longhttp://www.amazon.com/SHUTTLE-XS35V2-PC-Barebone-System/dp/B004XJCCQO...

bFlood

unread,
Dec 14, 2011, 10:56:48 AM12/14/11
to Google App Engine
exactly jeff, well put.

Rishi Arora

unread,
Dec 14, 2011, 11:03:45 AM12/14/11
to google-a...@googlegroups.com
Yeah, more RAM is linearly more costlier, which seems unfair and un-competitive.  It does make me wonder though, how hard does Google (and others like linode) try to actually make all that RAM available to you.  In other words, if you bought an instance with 4G RAM, do they absolutely guarantee you'll get all 4GB in physical RAM and won't start swapping because you're probably sharing the server with other apps?

exactly jeff, well put.
--
You received this message because you are subscribed to the Google Groups "Google App Engine" group.

Vivek Puri

unread,
Dec 14, 2011, 12:26:51 PM12/14/11
to Google App Engine
Great points Jeff. Double pricing for double CPU and memory is very
similar to Mac pricing on memory. So, what do new Mac buyers do? They
just get the lowest memory possible and get the cheapest memory deal
from ebay. Unfortunately, we cannot do that here.

On Dec 14, 10:36 am, Jeff Schnitzer <j...@infohazard.org> wrote:

Joshua Smith

unread,
Dec 14, 2011, 12:32:30 PM12/14/11
to google-a...@googlegroups.com
It is very likely that the servers don't have swap. I suspect that when google has a 1GB server, they stick just 7 128MB users on there. Remember: google's architecture is to have a zillion little computers, and distribute everything.

Jeff Schnitzer

unread,
Dec 14, 2011, 12:32:34 PM12/14/11
to google-a...@googlegroups.com
On Wed, Dec 14, 2011 at 1:26 PM, Vivek Puri <v...@vivekpuri.com> wrote:
> Great points Jeff. Double pricing for double CPU and memory is very
> similar to Mac pricing on memory. So, what do new Mac buyers do? They
> just get the lowest memory possible and get the cheapest memory deal
> from ebay. Unfortunately, we cannot do that here.

With other cloud providers and the Remote API you *totally* can do
that here. It's actually quite easy. It's just lame.

Jeff

Joshua Smith

unread,
Dec 14, 2011, 12:38:25 PM12/14/11
to google-a...@googlegroups.com
No, new Mac buyers get what the kid at the store tells them to get, and never open their mac, or buy memory from ebay.

If they were cheapskates, they'd be buying a PC that looks like a Mac on the outside, and costs a ton less.

Vivek Puri

unread,
Dec 14, 2011, 12:43:40 PM12/14/11
to Google App Engine
Saving your hard earned $$s doesnt equate to cheapskates. I just want
worth of my $$s that i earned after hours of toiling in front of a
screen. Sometimes 16 hours per day.

Vivek Puri

unread,
Dec 14, 2011, 12:48:23 PM12/14/11
to Google App Engine
I had hoped that AppEngine team will say - all of you guys are on F2
instances that cost $.08. I guess i was naive. So, all of us now by
default end up on these crappy instances that cannot run your code,
and pretty soon will be forced to upgrade to $.16 instances. Its very
similar to a worthless currency. Want to get 1 cup of coffee for 1
million YadaYadaDollars? Oh yeah, bring it on baby! Heck, even iPhone
1 came with same memory.

Vivek Puri

unread,
Dec 14, 2011, 1:00:33 PM12/14/11
to Google App Engine
Looking at Heroku, they offer instance for $.05 with 512mb RAM. Dont
see any wording asking for minimum hours of commitment. Besides that,
this minimum hours of commitment is such a pain. As i discovered, you
cannot increase/decrease hours at will. Any changes you make today, it
does not come into effect till next week. All these changes are
turning out to be very un-Google, and makes me feel i am dealing with
Verizon.

pdknsk

unread,
Dec 14, 2011, 1:32:06 PM12/14/11
to Google App Engine
You can get 8GB RAM for less than $40 now. Google probably pays $20.

John

unread,
Dec 14, 2011, 3:58:22 PM12/14/11
to google-a...@googlegroups.com
When it comes to writes I call it the "times two phenomenon". I have NO IDEA why you can't do anything that is not 2 writes???

e.g.
If you have a very simple Entity with 5 properties (none set to the NON default status of Unindexed) and you save it, it is 12 writes.


KeyWrite OpsID/NamefirstNamefourlastNamesixthree
agtwaXhvdG8tbGl2ZXILCxIEVGVzdBiNAQw12141JoedoorBobsticksfree

So you get slammed with 12 writes. Each property is 2 writes.


Here is a PropertyLess Entity
KeyWrite OpsID/Name
agtwaXhvdG8tbGl2ZXITCxIMUHJvcGVydHlMZXNzGI4BDA2142

2 writes.  Who needs properties anyhow?  That would mean you could query on them.  Queries return results, results are reads. Reads cost money.  

Oh wait, that is what memcache is for... wait a sec, memcache took down my whole site Monday from MemcacheServiceExceptions



"Also in my app app engine often spins up idle instances (that I cannot get rid of no matter what I configure) and send them exactly 1 request every 15 minutes so that I'm being charged the whole time for this instance I don't want."

This happens to me also... Why is it if you have 6 instances, 2 of them get most of the requests 3 of them get none and occasionally App Engine will start up a 7th instance while the idle 3 still get nothing? 


Don't get me wrong. I LOVE what App Engine stands for and I have all the respect in the world for the App Engine team.  BUT, I have been through SO much grief ranging from random app engine problems to having to migrate to an HR datastore to dramatic increases in pricing.  When I signed up for this (old pricing), I thought the pricing would eventually get better (almost like gmail and disk space), but instead it went the opposite.  Had my experience been perfect here and my app had run flawlessly all this time, I would have had no gripes and shut up and spent the extra cash without blinking.  But, instead I have experienced hair loosing problems, massive variations in performance and got stuck with a much larger bill.

John

unread,
Dec 14, 2011, 4:02:24 PM12/14/11
to google-a...@googlegroups.com
If computer time is "free", what is up with the exponential memory costs?

Joshua Smith

unread,
Dec 14, 2011, 4:07:54 PM12/14/11
to google-a...@googlegroups.com
Memory is a proxy for a fractional computer.

Also, go look up "exponential." You are using it wrong.

On Dec 14, 2011, at 4:02 PM, John wrote:

If computer time is "free", what is up with the exponential memory costs?

--
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/-/8QpGUUoyNJwJ.
Message has been deleted

André Pankraz

unread,
Dec 14, 2011, 4:21:08 PM12/14/11
to google-a...@googlegroups.com
the main problem with heroku seems to be that they start with a minimum of 200$ a month for the database - not very open source friendly?! it pays off if you use 1 TB data. Maybe I miss cheaper options there.

Joshua Smith

unread,
Dec 14, 2011, 4:36:13 PM12/14/11
to google-a...@googlegroups.com
So cost = K * size. That's linear, also called "Arithmetic" growth.  (It's not even "geometric" growth, which is what most people mean when they say "exponential" growth.)

On Dec 14, 2011, at 4:20 PM, John wrote:

Using it wrong?  Let's see.

My first choice is F1 instance with 128 MB ram  = .08/hr
If I want more, my next available option is F2 with 256 MB ram = .16/hr
If I want more, my only next available option is F4 with 512 MB ram = .32/hr

Each available choice is double the cost of the previous one.... hmmm.....

--
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/-/8Lsl5yaYtecJ.

Brandon Wirtz

unread,
Dec 14, 2011, 4:36:38 PM12/14/11
to google-a...@googlegroups.com

People are hung up on this 600mhz 128m Ram thing.  If you are using the API’s you are likely barely touching your CPU, and if you are using MemCache and Datastore most the time you aren’t using ram.

 

GAE is not the choice for Folding/Unfolding proteins or searching for ET. But if you are building Data Intense apps you can’t touch it on price.

--

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/-/NQ8l_6H79ZgJ.

André Pankraz

unread,
Dec 14, 2011, 4:50:44 PM12/14/11
to google-a...@googlegroups.com
so if we follow your image that hardware is free and we pay for the software licence this is quite similar to Oracles "paying for cores". we will see how good this K * size will work in the cloud environment ;)
sticking to your example: Oracle provides substantial sales discount that raises with the lump sum price.

mike hershey

unread,
Dec 14, 2011, 4:57:50 PM12/14/11
to google-a...@googlegroups.com
One last thing that is incredibly evil: paying customers do have an SLA, BUT if the SLA is not met its your responsibility to complain to Google to get credit towards your next month bill, and they don't have a metric of monthly uptime anywhere. To be able to claim against the SLA, I have to keep my own metrics of uptime.

Way to really stand behind your product Google. 

And if its not met:
    Monthly Uptime PercentagePercentage of monthly bill credited to future monthly bills of Customer
    99.00% – < 99.95%10%
    95.00% – < 99.00%25%
    < 95.00%50%


    But thats ok I'm sure they tell you who to contact:
    "To notify Google of SLA Financial Credit eligibility, please see the Documentation." 
- http://code.google.com/appengine/sla.html

I couldn't even imagine how mad I would be if my service was only up 95% of the time, and they only refunded 50% of my costs. 

    Its really just embarrassing. Get your shit together, get a real SLA that allows you to stand behind your product, lower the prices, and stop being evil. 

Joshua Smith

unread,
Dec 14, 2011, 4:58:52 PM12/14/11
to google-a...@googlegroups.com
As does google. Witness enterprise accounts (eliminating the $9 fees) and pre-paid instances. You gotta get pretty big before you start seeing discounts from any vendor.

On Dec 14, 2011, at 4:50 PM, André Pankraz wrote:

so if we follow your image that hardware is free and we pay for the software licence this is quite similar to Oracles "paying for cores". we will see how good this K * size will work in the cloud environment ;)
sticking to your example: Oracle provides substantial sales discount that raises with the lump sum price.

--
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/-/LGBgz84CWrgJ.

Jeff Schnitzer

unread,
Dec 14, 2011, 5:11:06 PM12/14/11
to google-a...@googlegroups.com
On Wed, Dec 14, 2011 at 5:36 PM, Brandon Wirtz <dra...@digerat.com> wrote:
> But if you are building Data Intense apps you can’t touch it on price.

Only if you can use the indexes provided. If you need a slightly
different index (say, a spatial index), you're forced to maintain it
in a third-party cloud. This was one of the original design goals for
Backends; I recall one of Ikai's posts describing a fulltext search
index as a use case. And yet backends are totally useless as index
repositories because they're priced 10X what it would cost to put the
index *anywhere* else.

1) You can't use backends as fast indexes because they are too expensive.
2) You can't use backends as persistent state because they aren't
reliable enough.

What can you use them for? They let you execute a single task longer
than 10minutes. Pretty weak sauce. They could have solved that
problem just by enabling long-running frontend requests url-by-url in
the app.yaml - that wouldn't require me to split my code and create
separate deployment modules.

I love Appengine, but Backends are a non-feature just like Email. It
would be better if Google engineers didn't waste their time creating
features nobody can use.

Jeff

Brandon Wirtz

unread,
Dec 14, 2011, 5:29:56 PM12/14/11
to google-a...@googlegroups.com
I have a spatial Index running on GAE using Calculated Tessellations as
indexed values. Based on a talk given by someone at Google About How they
optimized map searches for doing "with in radius" searches.

Sure we can't all be the genius architect I am (or possibly as good at
dissecting other people's information) but you trade what you store and how
you store it, in order to optimize for the platform.

But again it always comes back to people trying to make GAE act like other
platforms, It isn't. Is it better? Guess that depends on if you Like Ruby's
Philosophy of there are 10 ways to do everything, and not Wrong answers. Or
Python's There is only one way to do something and that way will be right.

GAE is about understanding what you need to do, and optimizing for the way
GAE wants you to do it. To Be honest I have never worked in a platform so
Rigid in architecture, or so limitless in potential.

I think "creative" problem solvers don't thrive on GAE. The rigidity stifles
them as they attempt to solve problems that don't need to be solved. And
Architects thrive because the Lego Pieces to play with are so abundant.


-----Original Message-----
From: google-a...@googlegroups.com
[mailto:google-a...@googlegroups.com] On Behalf Of Jeff Schnitzer
Sent: Wednesday, December 14, 2011 2:11 PM
To: google-a...@googlegroups.com
Subject: Re: [google-appengine] Re: Isn't .08/hr 1.92/day $59.52/month for a
600 MHZ CPU instance with 128 MB memory a LITTLE EXPENSIVE

Jeff

--


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

John

unread,
Dec 14, 2011, 5:39:43 PM12/14/11
to google-a...@googlegroups.com
As I stated "Each available choice is double the cost of the previous one".  That is exponential.  Am I missing something here??

e.g. 2^n where i is the index of the instance choice (0,1,2,...)

therefore

cost =  2^n * .08

Based on that, if an F8 were to be available, a 1024MB instance would be 2^3 *.08 = .64/hr

Jeff Schnitzer

unread,
Dec 14, 2011, 5:59:25 PM12/14/11
to google-a...@googlegroups.com
On Wed, Dec 14, 2011 at 6:29 PM, Brandon Wirtz <dra...@digerat.com> wrote:
> I have a spatial Index running on GAE using Calculated Tessellations as
> indexed values.  Based on a talk given by someone at Google About How they
> optimized map searches for doing "with in radius" searches.

Yeah yeah yeah, we can (and often do) come up with workarounds when
necessary. I use geohashing in a couple of my production apps. But
it these workarounds provide *very* narrow bounds around the problem
domain. One change to the sort, or one more inequality, and all bets
are off.

And that only works if your index is a well-known problem domain. I
was one of the early testers of Backends and used it for the index
that makes http://www.similarity.com/ run. I thought it was great.
Then Google announced pricing, and I quickly migrated the index to
rackspace cloud for one sixth the price.

I'm not saying there isn't always a workaround. But often that
workaround is "abandon GAE for part of your application". Of the four
major (and wildly-different) applications I've built on GAE, all have
required this "workaround". I'm pretty ok with that, except when the
only reason it's necessary is because of a bonkers pricing decision.

Jeff

Barry Hunter

unread,
Dec 14, 2011, 6:25:27 PM12/14/11
to google-a...@googlegroups.com
On Wed, Dec 14, 2011 at 10:39 PM, John <sc...@peoplepedia.org> wrote:
> As I stated "Each available choice is double the cost of the previous one".

Just because the memory/cpu just happens to be double too.

>  That is exponential.  Am I missing something here??

The size->price relationship is actually linear.

>
> e.g. 2^n where i is the index of the instance choice (0,1,2,...)
>
> therefore
>
> cost =  2^n * .08
>
> Based on that, if an F8 were to be available, a 1024MB instance would be 2^3
> *.08 = .64/hr

memory = 2^n * 128

too. Just coincidence.


If there was an F3 (n=3) with 384MB 1.8Gz, and $0.24 - it would no
longer be 'exponential' under your definiton.


>
> --
> 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/-/YsLOg3QQlAoJ.

Brandon Wirtz

unread,
Dec 14, 2011, 6:32:06 PM12/14/11
to google-a...@googlegroups.com
They aren't work around's they are Truth in computing. All the other
platforms have a layer between you and the data that is doing this same
thing. I might prefer to have a library but I like that I interface with my
data in a known way and understand what is happening "behind the scenes" and
can look at changes that are being made, or add my own optimizations.


-----Original Message-----
From: google-a...@googlegroups.com
[mailto:google-a...@googlegroups.com] On Behalf Of Jeff Schnitzer
Sent: Wednesday, December 14, 2011 2:59 PM
To: google-a...@googlegroups.com
Subject: Re: [google-appengine] Re: Isn't .08/hr 1.92/day $59.52/month for a
600 MHZ CPU instance with 128 MB memory a LITTLE EXPENSIVE

Jeff

--

mike hershey

unread,
Dec 14, 2011, 6:55:52 PM12/14/11
to google-a...@googlegroups.com
I think you've got that backwards. The other cloud services referenced (amazon/rackspace) are IaaS, and allow you to host your own operating system that you have complete control over. App engine is the service with a layer between the data and the application. You can't control how app engine datastore works. On rackspace/amazon you can host whatever database you want or make your own database. It its a totally different service model.

I'm not saying one is better then the other, but your implying that you have more control with app engine then you do with other cloud services when its quite the opposite.

I think most people who use app engine prefer it because you don't have to understand how everything works. I get a black box servlet environment, datastore, and whatever else a site typically needs. I don't have to waste time knowing how all of this works, I just get to use it. People who need the sort of control you are talking about tend to prefer IaaS services where you can control how everything works. 

Brandon Wirtz

unread,
Dec 14, 2011, 7:08:23 PM12/14/11
to google-a...@googlegroups.com

GAE FORCES you to think about your code. But it allows you to forget about everything else.

 

My implication was with in regards to Data handling.

 

I said GAE is rigid. It does what it does, and you can’t change it. 

 

BUT….

 

GAE is optimized to do Core things in the most optimal way possible. Google does indexed lookups on huge scale faster and cheaper than anything else.  If you want to index in some way Google doesn’t you have to implement that code in to the way Google Does Indexing.  When you run other software someone has written that code for you, often you can’t change it, you can’t mod it.

 

If You are working In the cloud you should be focusing on predictable scalable units that have a linear, or improved efficiency with scale.  Google Does this.  No one else does.  I can manage 100k instance software environment with GAE with a single developer.

 

Try that with any other platform, you can’t as you get bigger you will hit the limits of your Duplo blocks.  I can build the next Facebook on GAE. You can’t do that on Amazon. Because you will hit the back plain limits, the transaction limits, the ACID limits, the Elasticity is not Dynamic enough to handle changes in traffic hour by hour minute by minute.

 

GAE FORCES you to think about your code. But it allows you to forget about everything else.

 

 

From: google-a...@googlegroups.com [mailto:google-a...@googlegroups.com] On Behalf Of mike hershey
Sent: Wednesday, December 14, 2011 3:56 PM
To: google-a...@googlegroups.com
Subject: Re: [google-appengine] Re: Isn't .08/hr 1.92/day $59.52/month for a 600 MHZ CPU instance with 128 MB memory a LITTLE EXPENSIVE

 

I think you've got that backwards. The other cloud services referenced (amazon/rackspace) are IaaS, and allow you to host your own operating system that you have complete control over. App engine is the service with a layer between the data and the application. You can't control how app engine datastore works. On rackspace/amazon you can host whatever database you want or make your own database. It its a totally different service model.

 

I'm not saying one is better then the other, but your implying that you have more control with app engine then you do with other cloud services when its quite the opposite.

 

I think most people who use app engine prefer it because you don't have to understand how everything works. I get a black box servlet environment, datastore, and whatever else a site typically needs. I don't have to waste time knowing how all of this works, I just get to use it. People who need the sort of control you are talking about tend to prefer IaaS services where you can control how everything works. 

--

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/-/EvR7Ik_VjqIJ.

Jeff Schnitzer

unread,
Dec 14, 2011, 9:33:24 PM12/14/11
to google-a...@googlegroups.com
On Wed, Dec 14, 2011 at 8:08 PM, Brandon Wirtz <dra...@digerat.com> wrote:
> GAE FORCES you to think about your code. But it allows you to forget about
> everything else.

Brandon, I humbly suggest you just haven't hit an edge case yet.
There are plenty of indexing problems which GAE simply doesn't offer a
solution to. When I am forced to think about "does polygon A overlap
with polygon B?", I look for R-tree indexes... which GAE doesn't
offer. There are a million spatial index functions which are
no-brainers in PostGIS but represent man-years of work on GAE. And of
course there's fulltext indexing.

GAE gets more features every month, which is great. The magic
anti-exploding-index queries recently added are a godsend. Fulltext
indexing is on its way. And I'll be jumping up and down in happiness
when true spatial indexes show up. But let's not pretend that GAE is
complete. And let's make sure Google knows it when they make missteps
like pricing large instances unreasonably or offering halfway email
solutions that do little more than generate complaints on this list.

Jeff

Brandon Wirtz

unread,
Dec 14, 2011, 9:53:16 PM12/14/11
to google-a...@googlegroups.com
I have very nice Teselation Code, uses concentric Hexagons in place of
Circles and works for nested families on a map, and even has a "Driving"
distance function rather than "As the crow flies" that requires there be a
road of a given size in the Hex to make the move. And While it isn't well
tested we have a "driving Time" function that allows for City, Highway, and
Rural, hexes so that you can specify a that you'd like to find something
with in a 30 minute radius instead of a 30 mile radius, and if you can hop
on i80, that will search 30 miles, and if you are in LA it will search 7
miles. Currently only being used for US maps, but it can be used for Global.


Works really well and is blazing fast on GAE. Sell it to you for $500k.

-----Original Message-----
From: google-a...@googlegroups.com
[mailto:google-a...@googlegroups.com] On Behalf Of Jeff Schnitzer
Sent: Wednesday, December 14, 2011 6:33 PM
To: google-a...@googlegroups.com
Subject: Re: [google-appengine] Re: Isn't .08/hr 1.92/day $59.52/month for a
600 MHZ CPU instance with 128 MB memory a LITTLE EXPENSIVE

Jeff

--


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

Brandon Wirtz

unread,
Dec 14, 2011, 10:02:32 PM12/14/11
to google-a...@googlegroups.com
By the way the original use of the code was used for solving real life
traveling SalesMan routes for shipping and receiving for a VERY, VERY large
client, and GAE is not what the final product runs on but we needed
infrastructure to test the logic on which is what was sold, not the actual
source code.

Jeff Schnitzer

unread,
Dec 14, 2011, 11:22:36 PM12/14/11
to google-a...@googlegroups.com
If you have a polynomial-time implementation of optimal traveling
salesman, I'll buy that for $500k.

Jeff

--
We are the 20%

Bart Thate

unread,
Dec 14, 2011, 7:23:18 PM12/14/11
to google-a...@googlegroups.com
On Thu, Dec 15, 2011 at 12:55 AM, mike hershey <mikehe...@gmail.com> wrote:
I think most people who use app engine prefer it because you don't have to understand how everything works. I get a black box servlet environment, datastore, and whatever else a site typically needs. I don't have to waste time knowing how all of this works, I just get to use it. People who need the sort of control you are talking about tend to prefer IaaS services where you can control how everything works. 

You said the magic word .. "black box". In all these ramblings i never see the costs for the developers calculated in. GAE might be heaven for admins that dont want todo the dirty work themselves, for me GAE is a hell because i don't have any control over the runtime enironment at all. If something is not working i can't check logs, i have to buy a 500 dollar a month package to get some support from Google to go look what is going wrong. Nothing worse then having your app misbehave, shout something in the IRC channel and after some time, heee its fixed ! Gives me no clue of whats going on and just that is so important for me when i develop things. 

The other thing that connects to the developers costs thing i have against GAE and more against Google is the use of API. These API get deprecated quicker then i can start reading their docs, and when they get into "mainstream" i have to fork over money to use them. As i write software for other people to use, that would require them to make use of this *paid* API so i can never build upon them in the core of my app. Sure i  can make a plugin for this API and make it optional, but say like integrating the translation API into my bot (would be extreeemly usefull) is a nono. We need FOA .. Free and Open API, things i can rely on. Not so with Google though.

Last thing is that Google choose the evil path, it has killed the hippie amongst all of us programmers. The vibrant community that was created with Wave has also disappeared with the death of it. Is the HRD the holy grail of cloud computing (only reason i would use GAE i think), it is there because so many of us went down the rough road on M/S in those early days. Now i feel the spirit is lost and people are only choosing GAE because they have already so much invested in it (developer wise).

I miss the socket on GAE, really do. Programming with sockets is so much better because you get "pure" internet not just HTTP and streaming is possible. Streaming people ! Stream it ;]
 
Bart Thate

programming schizophrenic - "till freedom come!"


Brandon Wirtz

unread,
Dec 15, 2011, 1:49:38 AM12/15/11
to google-a...@googlegroups.com
Optimal is a strong word.

12 trucks have to deliver 144 items and pick up 60 items, No truck can have
more than 14 items at any given time. Do to a Quirk in what we are shipping
let's say it is dogs and cats... You can never be hauling both a dog, and a
cat at the same time, but you can be picking up a dog, at a location you
just deposited a cat. Also if you travel with a Dog for more than 3 hours
you explode, but you can trade dogs for ones that won't explode at any place
that can swap dogs with you.

While I call this a traveling sales man, we also have an acceptable penalty
for failed pickups and failed deliveries, so it is more chess engine than
Traveling Salesman. Capture X pieces in Y moves and score as many points
as possible.

My piece was really just building the What are the valid moves at this point
in time. I didn't have to solve all the rest.

Also The items being shipped were not Dogs, They were something far more
Fissile.

Andrei

unread,
Dec 15, 2011, 1:54:24 AM12/15/11
to Google App Engine
Goog only very recently switched to new pricing and they need to see
how it sells
It is probably will not sell good and significant number of people
will leave
So after they figure it out they'll make it cheaper, they'll have to
It will probably take at least half year for them to figure current
pricing model is bad

I know it does not sound good, but they better do same as AWS for
pricing
Make trial for something like 6 month and after that no more free apps
This way prices spread more evenly for everybody and eliminate stupid
9/month fee
But only time will show

Also why can not they just sell virtual servers like AWS

On Dec 14, 6:50 am, John <sc...@peoplepedia.org> wrote:
> Just started thinking about this... but now that we are basically paying
> for all the datastore reads/writes, and bandwidth separately...
>
> Isn't paying $60 a month for a 600 MHZ instance with only 128 MB ram a
> little expensive?
>
> Just taking a quick glance at EBAY, I can buy a > 2 GHZ machines with over
> a GB of memory all day long.
> I can buy BRAND NEW Intel Atom Dual-Core D525 Processor(1.8GHz, 1MB L2
> Cache), Support Intel Hyper-Threading technology,
> with 1GB memory for ~ $160 all day longhttp://www.amazon.com/SHUTTLE-XS35V2-PC-Barebone-System/dp/B004XJCCQO...
>
> Call me crazy, but I still have my 1 GHZ pc I bought back in 1999 (12 years
> ago) sitting in the garage and I would have a problem giving it away (It
> also has a lot more memory than 128 MB ram).
>
> A standard (small) SAME PRICEd Amazon EC2 instance comes with 1.7 GB of
> memory and even their FREE micro instance gives you 613 MB of memory.
>
> I understand computers were a lot more expensive back in 1999, but they
> have gotten a lot cheaper over the past few years.
>
> Please justify what I am paying for because right now I am trying to
> justify upgrading to the F2 instance class for twice the price ($120/month)
> just so I can double up and get a whopping 256MB ram!

jon

unread,
Dec 15, 2011, 6:51:54 AM12/15/11
to Google App Engine
John, by default MemcacheServiceException should not be thrown and
take down anyone's site. It's a bug (that has affected my site in the
past too).

I've filed a bug report here:
http://code.google.com/p/googleappengine/issues/detail?id=6236

On Dec 15, 7:58 am, John <sc...@peoplepedia.org> wrote:
> When it comes to writes I call it the "times two phenomenon". I have NO
> IDEA why you can't do anything that is not 2 writes???
>
> e.g.
> If you have a very simple Entity with 5 properties (none set to the NON
> default status of Unindexed) and you save it, it is 12 writes.
>

> Key <http://localhost:8080/_ah/admin/datastore?&kind=Test&order=__key__>Write
> OpsID/Name<http://localhost:8080/_ah/admin/datastore?&kind=Test&order=__key__>
> firstName<http://localhost:8080/_ah/admin/datastore?&kind=Test&order=firstName>
> four <http://localhost:8080/_ah/admin/datastore?&kind=Test&order=four>
> lastName<http://localhost:8080/_ah/admin/datastore?&kind=Test&order=lastName>
> six <http://localhost:8080/_ah/admin/datastore?&kind=Test&order=six>three<http://localhost:8080/_ah/admin/datastore?&kind=Test&order=three>
> agtwaXhvdG8tbGl2ZXILCxIEVGVzdBiNAQw12141JoedoorBobsticksfree


>
> So you get slammed with 12 writes. Each property is 2 writes.
>
> Here is a PropertyLess Entity

> Key<http://localhost:8080/_ah/admin/datastore?&kind=PropertyLess&order=-_...>Write
> OpsID/Name<http://localhost:8080/_ah/admin/datastore?&kind=PropertyLess&order=-_...>
> agtwaXhvdG8tbGl2ZXITCxIMUHJvcGVydHlMZXNzGI4BDA2142


>
> 2 writes.  Who needs properties anyhow?  That would mean you could query on
> them.  Queries return results, results are reads. Reads cost money.
>
> Oh wait, that is what memcache is for... wait a sec, memcache took down my
> whole site Monday from MemcacheServiceExceptions
>

> http://code.google.com/status/appengine/detail/memcache/2011/12/12#ae...

Brian Quinlan

unread,
Dec 15, 2011, 7:12:05 AM12/15/11
to google-a...@googlegroups.com
HI John,

On Thu, Dec 15, 2011 at 7:58 AM, John <sc...@peoplepedia.org> wrote:
When it comes to writes I call it the "times two phenomenon". I have NO IDEA why you can't do anything that is not 2 writes???

I'm not sure if this is what you are looking for but the translation of high-level datastore operations into billable operations is described here:

Cheers,
Brian
 

e.g.
If you have a very simple Entity with 5 properties (none set to the NON default status of Unindexed) and you save it, it is 12 writes.


Key Write Ops ID/Name firstName four lastName six three
agtwaXhvdG8tbGl2ZXILCxIEVGVzdBiNAQw 12 141 Joe door Bob sticks free

So you get slammed with 12 writes. Each property is 2 writes.


Here is a PropertyLess Entity
Key Write Ops ID/Name
agtwaXhvdG8tbGl2ZXITCxIMUHJvcGVydHlMZXNzGI4BDA 2142

2 writes.  Who needs properties anyhow?  That would mean you could query on them.  Queries return results, results are reads. Reads cost money.  

Oh wait, that is what memcache is for... wait a sec, memcache took down my whole site Monday from MemcacheServiceExceptions



"Also in my app app engine often spins up idle instances (that I cannot get rid of no matter what I configure) and send them exactly 1 request every 15 minutes so that I'm being charged the whole time for this instance I don't want."

This happens to me also... Why is it if you have 6 instances, 2 of them get most of the requests 3 of them get none and occasionally App Engine will start up a 7th instance while the idle 3 still get nothing? 


Don't get me wrong. I LOVE what App Engine stands for and I have all the respect in the world for the App Engine team.  BUT, I have been through SO much grief ranging from random app engine problems to having to migrate to an HR datastore to dramatic increases in pricing.  When I signed up for this (old pricing), I thought the pricing would eventually get better (almost like gmail and disk space), but instead it went the opposite.  Had my experience been perfect here and my app had run flawlessly all this time, I would have had no gripes and shut up and spent the extra cash without blinking.  But, instead I have experienced hair loosing problems, massive variations in performance and got stuck with a much larger bill.

--
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/-/DElz4NPSc8oJ.

JH

unread,
Dec 15, 2011, 8:47:39 AM12/15/11
to Google App Engine
Actually datastore queries can use quite a bit of ram

On Dec 14, 3:36 pm, "Brandon Wirtz" <drak...@digerat.com> wrote:
> People are hung up on this 600mhz 128m Ram thing.  If you are using the
> API’s you are likely barely touching your CPU, and if you are using MemCache
> and Datastore most the time you aren’t using ram.
>
> GAE is not the choice for Folding/Unfolding proteins or searching for ET.
> But if you are building Data Intense apps you can’t touch it on price.
>
> From: google-a...@googlegroups.com
> [mailto:google-a...@googlegroups.com] On Behalf Of André Pankraz
> Sent: Wednesday, December 14, 2011 1:21 PM
> To: google-a...@googlegroups.com
> Subject: [google-appengine] Re: Isn't .08/hr 1.92/day $59.52/month for a 600
> MHZ CPU instance with 128 MB memory a LITTLE EXPENSIVE
>
> the main problem with heroku seems to be that they start with a minimum of
> 200$ a month for the database - not very open source friendly?! it pays off
> if you use 1 TB data. Maybe I miss cheaper options there.
>
> --
> You received this message because you are subscribed to the Google Groups
> "Google App Engine" group.

> To view this discussion on the web visithttps://groups.google.com/d/msg/google-appengine/-/NQ8l_6H79ZgJ.

Message has been deleted

mike hershey

unread,
Dec 15, 2011, 2:07:17 PM12/15/11
to google-a...@googlegroups.com
Woops, better take down those pictures! Your violating adsense's ToS. Your not allowed to share CPM with anyone. Just a heads up, I would hate to see your adsense account disabled

Brandon Wirtz

unread,
Dec 15, 2011, 2:13:20 PM12/15/11
to google-a...@googlegroups.com

That isn’t actually the rule.

And if they kick him out for it I’ll loan him my lawyer. 

 

AppEngine is great for SEO, and likely your pages load faster running on GAE which is good for CTR.

 

 

 

From: google-a...@googlegroups.com [mailto:google-a...@googlegroups.com] On Behalf Of mike hershey
Sent: Thursday, December 15, 2011 11:07 AM
To: google-a...@googlegroups.com
Subject: [google-appengine] Re: Isn't .08/hr 1.92/day $59.52/month for a 600 MHZ CPU instance with 128 MB memory a LITTLE EXPENSIVE

 

Woops, better take down those pictures! Your violating adsense's ToS. Your not allowed to share CPM with anyone. Just a heads up, I would hate to see your adsense account disabled

--

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/-/xCXOCy-d6FcJ.

mike hershey

unread,
Dec 15, 2011, 2:18:08 PM12/15/11
to google-a...@googlegroups.com
"7.      Confidentiality. You agree not to disclose Google Confidential Information without Google's prior written consent. "Google Confidential Information" includes without limitation: (a) all Google software, technology, programming, specifications, materials, guidelines and documentation relating to the Program; (b) click-through rates or other statistics relating to Property performance in the Program provided to You by Google; and (c) any other information designated in writing by Google as "Confidential" or an equivalent designation. However, You may accurately disclose the amount of Google’s gross payments to You pursuant to the Program. Google Confidential Information does not include information that has become publicly known through no breach by You or Google, or information that has been (i) independently developed without access to Google Confidential Information, as evidenced in writing; (ii) rightfully received by You from a third party; or (iii) required to be disclosed by law or by a governmental authority."

Brandon Wirtz

unread,
Dec 15, 2011, 2:51:56 PM12/15/11
to google-a...@googlegroups.com

Not the forum for discussing, but you can share with your Affiliates, which has no definition.  And I would say you are my affiliate.  Also there are several California cases (which govern this contract) that say that a contract cannot require that parties don’t disclose the price of goods, as part of anti-monopoly requirements.

 

If you want to look at the contract, Clause 14 says that if Google is found to be a monopoly in the Advertising or search space, you are required to pay for their defense.

 

Paragraph 17 says the contract is bound to California law, unless California law conflicts with the contract terms.

 

Paragraph 3 precludes you ever talking to any advertiser who you have written a review about if they have advertised on your site through Adsense…

 

Paragraph 5 says as an adsense user you are never allowed to tell someone to “Google …. “ the topic.

 

Clause 8 says that if Google doesn’t want to pay you they don’t have to.

 

If you think that the Adsense TOS is enforced as written you are mistaken.  They will let you do anything you want if you make them enough money, and they will screw you over if it suits them, or if they aren’t paying attention and it is inconvenient to fix.

--

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/-/b6rwZei2eXMJ.

Reply all
Reply to author
Forward
0 new messages