63% Reduction in Cold Start Time by Changing AppID

[Kyle Finley]

Has anyone noticed a dramatic difference in Cold Start times when using different AppIDs?

I recently changed the AppID for a project that I'm working on and there was a noticeable reduction in the cold start time.

Intrigued, I decided to perform an experiment:

• I loaded the application on 4 different AppIDs all of which were HRD.
  
• Using tools.pingdom.com I loaded the home page for each of the AppIDs using the http://*.appspot.com domain and recorded wait (yellow bar) time for the HTML page.
• I, then, shutdown each of the instance and repeated the test.

In this limited sample the fasted AppID loaded the application in an average of 1.35 secs, while the slowest averaged 3.55 secs

The Application is using:

  runtime: python27

  threadsafe: false

I also performed the test on warm instances, but the results did not seem to indicate a correlation.

I also tested the fasted and slowest AppIDs with a similar application written in Go, where AppID did not effect the cold start times.

A little more information on the AppIDs:

• The fasted AppID was created yesterday
• The slowest was created many months ago.
  
• The slowest AppIDs were also the ones containing a dash(-), but that's probably just a coincidence.

Is this slowness due to older hardware, or is this somehow related to the python27 runtime?

Can anything be done for these slow AppIDs? Is it best just to delete them?

Thanks,

- Kyle

https://docs.google.com/spreadsheet/ccc?key=0Apr2o_vtfYo3dFNPMDQ2ZHU1T2ZwcmJIOHNQYU9EWXc

[James Broberg]

Did you check the current memory utilisation of each app when you ran the test?

> --
> 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/-/CH89M2Tb9ekJ.
> 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.

[Kyle Finley]

I don't record it on each test, but it was generally ~ 42.2 MBytes +-0.4