Google App Engine Load Test Result

Posted by Niclas Nilsson on Jun 05, 2008

Sections

Architecture & Design,

Development,

Operations & Infrastructure

Topics

Cloud Computing ,

Architecture ,

Web Frameworks ,

Software Testing ,

Java ,

Rich Internet Apps

Tags

GWT ,

Google ,

PaaS ,

Performance Evaluation

As we reported previously, there was a public Google Web Toolkit on Google App Engine load test recently. The result of the test was that the application got an average of 10 reqs/sec during one hour, and a peak load of 35 reqs/sec. In total, the application received 40 000 requests in one hour.

According to Didier Girard who set up the test, the users couldn’t notice the load.

Didier also did another thing to make the situation more like the real thing. He upgraded the software during the test.

Just before 4PM, I put a new version of the application in production. At that time their was about 10 requests/second. It was unnecessary and risky. But sometimes you have to upload a new version of your application during high traffic and I wanted to see if it is possible with GAE. All run perfectly.

Pete Koomen, Product Manager of Google Application Engine gave Didier some hints on how to help the application scale. In short:

• Avoid unbounded queries.
• Do computation incrementally using the datastore, rather than all at once.
• Avoid contention on datastore entities.
• Avoid large entity groups.

Close to 40% of the traffic came from InfoQ readers, even though the article was published less than an hour before the test. Didier says that he’s is planning another round of load testing, so those who would like to participate in further test runs should keep an eye on Tracking News on GWT.

1. Back to top

   damn

   by Christoph Blank

   we need more users then! :)
   
   good news though

   Reply

2. Back to top

   10?

   by Cameron Purdy

   Ten requests per second? Forty thousand in an hour? Come on .. get forty thousand per second and it'll be interesting.
   
   
   
   (And yes, there are sites that sustain that order of magnitude load .. on the order of billions of pages served each day.)
   
   
   
   Peace,
   
   
   
   Cameron Purdy
   
   Oracle Coherence: Data Grid for Java, .NET and C++

   Reply

3. Back to top

   Re: 10?

   by Niclas Nilsson

   But there are still many websites that would love to have 40K hits per hour.
   

   
   My impression from his blog posts was not that he tried to prove that it works for the extreme, fraction of a percent, high-traffic websites - just that a quite naive (Knuth-style) GWT/GAE app can work quite well without much effort for 99.x% (?) of all webapps out there.
   

   
   Kind regards
   
   Niclas

   Reply

4. Back to top

   Requests per second need context

   by Slava Imeshev

   Requests per second need context. You are talking about the server side of the process without considering response time and request back-log.
   
   
   
   Also, performance only makes sense in the context of requirements. The 10 request/sec can be really a lot of that app have only been able to process 1 request/per and it is abysmally horrible if we are talking sites such as Yahoo, eBay or InfoQ.
   
   
   
   Another thing to consider is scalability - will your application be able 1000 requests per second if you add 100 Google App Engine?
   
   
   
   
   Without this incomplete list of considerations 10 requests/sec are as good as the answer to life, the universe, and everything.
   
   
   
   
   Hope this helps.
   
   
   
   
   Regards,
   
   
   
   
   Slava Imeshev
   
   
   Cacheonix - Clustered Cache and Data Grid for Java
   
   
   

   Reply

• Older >