CloudFlare's new web protocol speeds up dynamic web performance
blog.cloudflare.com
blog.cloudflare.com
After a couple of days of trial and error (kept failing due to too many open file handles) we now have it running smoothly but have noticed little to no speed improvement both vs what it was before and vs other sites on the same environment.
Would be interesting to know if anyone else has had a similar/dissimilar experience...?
Homework done, measurements taken. Happy with the results.
The only thing that we saw a slowdown for was when we connect to pages from our office, when our servers are in town.. but that's not the fault of Cloudflare, but simply that they don't have servers in the same city as us!
$ dig blog.cloudflare.com
...
;; ANSWER SECTION:
blog.cloudflare.com. 21 IN CNAME blog.cloudflare.com.cdn.cloudflare.net.
blog.cloudflare.com.cdn.cloudflare.net. 21 IN CNAME cloudflare.posterous.com.
cloudflare.posterous.com. 1243 IN A 184.106.20.99
So it looks like requests are going directly to Posterous. Seems like an obvious place for CloudFlare to showcase their speed.Every other major CDN offers some variation on this same service. Suck the bits in at a server close to the origin, compress them using proprietary voodoo, expand things at the server closest to the client.
Akamai owns the relevant patents on this (which it acquired from the Netli acquisition) and has used them to sue then acquire other companies like Cotendo. See http://www.google.com/patents/US6820133 and http://www.google.com/patents/US7293093
The obvious win is pushing less bytes across the wire. The less obvious win is that it allows the application server response to complete VERY fast, freeing up the app server thread to handle more requests. This is the larger win for large scale 3-tier applications.