I agree this is going to be an issue for applications that are not read-heavy. The more interactivity and state changing done by users in your app, the more expensive it is going to be to host on GAE.
Googlers have hinted that they're working on getting the CPU-cost for DS operations down, but it's likely to be shaving some % off more than anything drastic. On Feb 16, 9:01 pm, diomedes <[email protected]> wrote: > Ok, > > (a bit bummed that nobody bother to reply to my first post to the > group) > I spent the last few days reading related posts, transcripts of chats > with Marzia, went through the related tickets and still I have no > answer. > > Let me try once more to explain my question: > It is not about quotas. > It is neither about performance per se. > My question is about cost. > My app implements a beacon service - my client websites that will be > using the beacon have 5-10M pageviews per month and these pageviews > will result into actual beacon hits to my app. > These requests do not hit the Datastore - I buffer them in memcached > doing minimal processing to keep the per beacon-hit cost low and > "process them" in batches every few seconds. > > Still, in spite of all the buffering, if the cheapest write (1 attr > table, no indexes) gets charged 250-500msec, this make the app design > for such a high throughput service non-obvious. > > I understand that I can decrease the number of my writes by using a > pickled blob attr that contains lots of records inside - that what I > am about to do. > I just wanted a confirmation from any gurus out there, or from the > google team that my understanding about the cost of the "cheapest > write" is correct. > > I tried batch db.put( of 100 single cell objects) - still took 23 > seconds of data store CPU , i.e. my cheapest write = 230msec - a bit > less than before but not by much. > I made all 100 objects children of same parent and performed the same > test - again same datastore CPU utilization > > Is there another method to update/insert multiple records that costs > less? I.e. is there any cheaper write? ( I am very flexible :-) ) > Is my understanding that the planned 10-12cents per CPU hr will be > applied towards the datastore CPU usage as well? > > Thanks a lot, > > diomedes > > On Feb 14, 3:18 pm, diomedes <[email protected]> wrote: > > > Hi all, > > > I have a question with regards to the pricing and how it relates to > > the relatively high API CPU associated with each write (400-500msec) > > > I have started working on my first GAE app about a month ago - and > > overall I am both excited and very satisfied. My initial plan, was to > > run it on Amazon's EC2 - but eventually I took the plunge, started > > learning python, move to GAE and (almost) never looked back :-) > > > My app, a cacti-like webservice that monitors the performance ( think > > response time) of a website using google analytics-like beacons, is > > rather resource demanding. On top of that GAE best practices imply > > that any expensive reports/aggregates etc should be precalculated/ > > stored instead of dynamically produced on demand. All that result in > > many writes and given that the simplest write (single key-val pair, no > > indexes) gets "charged" approx 500msec of API cpu time (see related > > thread by > > Matijahttp://groups.google.com/group/google-appengine/browse_thread/thread/...) > > a normal DB design that would have been meaningful in terms of cost on > > EC2 becomes impossible on GAE. > > > Because I am a google-aholic I decided to change the app design to > > minimize writes - I fetch a bunch of pickled data as a blob, update > > in mem and write them back as blob (just like people did before DBs > > came along :-) ) > > Before I commit to that design I wanted to get the confirmation that > > my understanding is correct: > > - Google is going to charge 10-12cents per CPU-hour and it will > > include in that all the CPU used from APIs etc. (http:// > > googleappengine.blogspot.com/2008/05/announcing-open-signups- > > expected.html) > > - This means that if your site does 10M pageviews a month and does > > a couple writes per pageview at 500msec per write it will be "10M CPU > > secs/mo just from the writes, i.e. 10M/3600 * $.10/hr = $280/mo just > > from the writes. > > > Is this correct? > > > For the record, I find Google's planned pricing extremely attractive > > when compared to Amazon's primarily due to the fact that Amazon > > charges 10c for CPU-hr of the machine while google (will) charge 10c > > for CPU-hr *actually used* by your requests. This makes a huge > > difference -- a server running at 50+% capacity (thats rather > > aggressive - but with Amazon/RightScale combination you can be > > aggressive) will still use less than 20% of its CPU during that time. > > However, when comparing the cost writes between Google and the > > corresponding setup of a [high CPU EC2 server + elastic storage] combo > > (able to provide quite more than 20-50 "simple" writes per sec) > > Amazon is much cheaper than Google. > > > Ok, that's all I had to say, > > Sorry for the rather long post, > > Looking forward to hear comments > > > Ah and thank you very very much for lifting the high cpu quota!! > > > Diomedes --~--~---------~--~----~------------~-------~--~----~ 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 -~----------~----~----~----~------~----~------~--~---
