HipHop for PHP in Production at Hyves

Re:HipHop for PHP in Production at Hyves by ck2 on Thursday December 22, 2011 @10:43AM (#3381845)

How does hiphop performance compare to fastcgi + opcode-cache (with stat turned off)

I remember reading that on paper it seems it should be faster but in reality not so much.

Ah okay, hiphop IS faster:

http://php.webtutor.pl/en/2011/05/17/drupal-hiphop-for-php-v...

   environment [req/sec][ms/req][%]
   Regular PHP 	15.10 	66.242 	100%
   PHP + APC 	46.90 	21.321 	310%
   HipHop  	77.01 	12.985 	510%  (1.6x faster than apc)

http://huichen.org/en/2010/06/wordpress-3-benchmark/

   wp3+php+apc  30.86   32.400  
   wp3+HipHop   43.50   22.990        (1.4x faster than apc)

Maybe multicast could be used on the intranet for distribution - but 500mb monolithic binaries sounds kinda insane.

[ Parent | on HN ]

Re:HipHop for PHP in Production at Hyves by patangay on Thursday December 22, 2011 @11:15AM (#3381993)

I work for facebook and I worked on the deployment of these large hiphop binaries of our code base. We considered multiple options - binary diffs, multicast, etc. The problem with multicast is that it's hard to configure and maintain in our complex cross region datacenter setup that has to travel through other peering networks.

We ended up with a torrent deployment system that scales beautifully.

[ Parent | on HN ]

Re:HipHop for PHP in Production at Hyves by Ramonster on Tuesday January 03, 2012 @03:35AM (#3419039)

We ended up doing exactly the same, I wrote a tracker for the bittorrent deploy system that is aware of our network topology and able to transport these 500Mb binaries to our frontend pool in < 3mins (300+ servers)

The tracker is opensource: https://github.com/hyves-org/p2ptracker

We're working on open sourcing more moving parts of our deploypipeline.

I'd be interested to compare notes on your system ?

[ Parent | on HN ]