5 ms·

Did a quick comparison with my own service, BitBalloon. On appengine you're just deploying a dynamic app that just routes everything to a static folder, but si

by bobfunk 13y ago

Did a quick comparison with my own service, BitBalloon. On appengine you're just deploying a dynamic app that just routes everything to a static folder, but since Google doesn't know this is a static site, it's pretty limited what they can do to set good cache headers and optimize stuff for performance. So even if they have an awesome infrastructure, BitBalloon will make your site perform better. Here's the quick test result from the same site uploaded to AppEngine and BitBalloon: http://tools.pingdom.com/fpt/#!/vLi9d/http://teststaticsite.appspot.com/ http://tools.pingdom.com/fpt/#!/vLi9d/http://teststaticsite.... http://tools.pingdom.com/fpt/#!/dwqwvX/http://speedtest.bitballoon.com http://tools.pingdom.com/fpt/#!/dwqwvX/http://speedtest.bitb...

StavrosK 13y ago
That's a good observation. I wrote a small open-source script to do the same thing: https://github.com/stochastic-technologies/static-appengine-hoster https://github.com/stochastic-technologies/static-appengine-... It supports multiple domains for a single AppEngine instance, and custom URLs for each page. Pull requests for caching/headers/etc welcome!
ithkuil 13y ago
https://developers.google.com/appengine/docs/python/config/appconfig#Python_app_yaml_Static_cache_expiration https://developers.google.com/appengine/docs/python/config/a...
scottfr 13y ago
I believe App Engine actually has specific handling for files you identify as being static. This bypasses much of the dynamic overhead.
codeka 13y ago
The difference is much lower if you run the App Engine request a second time. The first time you run it, it needs to spin up an instance which can take some time. If I run it a second time I get 770ms for bitballoon and 955ms for App Engine, which is much less significant. Also, running the test from my local computer, app engine wins out presumably because Google's network is basically unmatched when it comes to proximity to your users. Additionally, if you actually define your static files as static, then it doesn't even need to spin up instances to serve them. I've certainly had my criticisms of App Engine in the past, and for certain workloads it's not appropriate [1], but for mostly read only websites like blogs, news, landing pages, I even run a forum on app engine, it's perfect. In fact, that blog post where I complain about the cost of app engine for hosting my game is itself hosted on app engine. It cost me 6c the day it hit the front page of hn. I pay more for the domain than I do for app engine hosting. [1]: http://www.war-worlds.com/blog/2013/06/switched-away-from-app-engine-couldnt-be-happier http://www.war-worlds.com/blog/2013/06/switched-away-from-ap...
adam74 13y ago
The speed difference is negligible. The price difference for a custom domain is not.
- bobfunk 13y ago
  Peoples definition of negligible probably varies. Half a second is definitively enough that plenty of large scale tests have shown differences in conversion rate. That difference gets much bigger for people with more than 1 page view due to perfect caching headers.

by bobfunk 13y ago

StavrosK 13y ago

That's a good observation. I wrote a small open-source script to do the same thing: https://github.com/stochastic-technologies/static-appengine-hoster https://github.com/stochastic-technologies/static-appengine-... It supports multiple domains for a single AppEngine instance, and custom URLs for each page. Pull requests for caching/headers/etc welcome!

ithkuil 13y ago

https://developers.google.com/appengine/docs/python/config/appconfig#Python_app_yaml_Static_cache_expiration https://developers.google.com/appengine/docs/python/config/a...

scottfr 13y ago

I believe App Engine actually has specific handling for files you identify as being static. This bypasses much of the dynamic overhead.

codeka 13y ago

The difference is much lower if you run the App Engine request a second time. The first time you run it, it needs to spin up an instance which can take some time. If I run it a second time I get 770ms for bitballoon and 955ms for App Engine, which is much less significant. Also, running the test from my local computer, app engine wins out presumably because Google's network is basically unmatched when it comes to proximity to your users. Additionally, if you actually define your static files as static, then it doesn't even need to spin up instances to serve them. I've certainly had my criticisms of App Engine in the past, and for certain workloads it's not appropriate [1], but for mostly read only websites like blogs, news, landing pages, I even run a forum on app engine, it's perfect. In fact, that blog post where I complain about the cost of app engine for hosting my game is itself hosted on app engine. It cost me 6c the day it hit the front page of hn. I pay more for the domain than I do for app engine hosting. [1]: http://www.war-worlds.com/blog/2013/06/switched-away-from-app-engine-couldnt-be-happier http://www.war-worlds.com/blog/2013/06/switched-away-from-ap...

adam74 13y ago

The speed difference is negligible. The price difference for a custom domain is not.

bobfunk 13y ago
Peoples definition of negligible probably varies. Half a second is definitively enough that plenty of large scale tests have shown differences in conversion rate. That difference gets much bigger for people with more than 1 page view due to perfect caching headers.