Winpdb - A Platform Independent Python Debugger

Some people are under the impression that the Google App Engine (GAE) will allow their application to scale as big as earth. However, if you read their quota document you will notice there are limits even for billing-enabled accounts which limit GAE to the scale of about 1 or 2 dedicated servers.

The CPU usage is limited to 72 CPU minutes per minute. A Google CPU is defined as 1.2GHz x86 CPU and since they are famous for using low cost commodity servers (probably focusing on energy efficiency) I estimate these 1.2GHz CPU are equivalent to 0.6GHz modern E54xx Xeon core and that therefore 72 Google CPU are roughly the equivalent of 2 modern 2.5GHz Xeon machines.

The service is also limited to about 500 requests (hits, not page views) per second which a single dedicated server should be able to handle. Google offers a form through which you can request additional quota, but I would not recommend anyone to base their business plan on begging Google for that extra quota.

Maybe the Google App Engine can not safely scale any bigger for an arbitrary application. It is one thing to distribute a well designed algorithm using the MapReduce methodology over 1,000 servers, but it is another thing to distribute an arbitrary application. For scalability you ideally want an application designed according to the shared-nothing principle (http://en.wikipedia.org/wiki/Shared_nothing_architecture) and what is more shared-nothing than two different and separate small web applications? Keeping these applications small allows Google to distribute their processing efficiently.

References
http://code.google.com/appengine/docs/quotas.html
http://code.google.com/appengine/docs/billing.html
http://news.cnet.com/8301-1001_3-10209580-92.html
http://www.gigabyte.com.tw/Products/Networking/Products_Spec.aspx?ProductID=1075
http://elnexus.com/articles/2004-2008.aspx
http://2bits.com/articles/can-a-drupal-web-site-handle-a-million-page-views-a-day.html