Speeding up PHP with the HipHop VM

Things make sense now. I remember a while back Facebook invested in an experiment with PHP on PyPy. They didn't pursue it, even though it produced impressive results. It seems their own in-house JIT has better performance?

Re: Speeding up PHP with the HipHop VM

#6

Post by fomojola » Thu, Nov 29, 2012, 10:26 PM UTC

I'm curious if they were still evolving HPHPc at the same time as they were evolving HHVM: the chart shows HPHPc as having flat performance over time, while HHVM was getting better over a 7-month period. Could HPHPc have achieved the same performance gains if the same effort was expended?

Re: Speeding up PHP with the HipHop VM

#7

Post by kmavm » Thu, Nov 29, 2012, 10:30 PM UTC

fomojola wrote: post #6

I'm curious if they were still evolving HPHPc at the same time as they were evolving HHVM: the chart shows HPHPc as having flat performance over time, while HHVM was getting better over a 7-month period. Could HPHPc have achieved the same performance gains if the same effort was expended?

Hey, I'm an HHVM engineer.

We just normalized HPHPc performance to make the graph easier to read. HPHPc actually got considerably faster over the same period as well; since both systems share a runtime, many changes helped both.

HPHPc probably got about 20% faster over 2012, even though nobody was actively working on it, mostly through happy side effects of work that was directed at HHVM.

Re: Speeding up PHP with the HipHop VM

#8

Post by olaf » Thu, Nov 29, 2012, 10:31 PM UTC

"So, when you combine XHP with HipHop PHP you can start to imagine that the performance penalty would be a lot less than 75% and it becomes a viable approach." http://toys.lerdorf.com/archives/54-A-quick-look-at-XHP.html

Re: Speeding up PHP with the HipHop VM

#9

Post by nbm » Thu, Nov 29, 2012, 10:34 PM UTC

fomojola wrote: post #6

I'm curious if they were still evolving HPHPc at the same time as they were evolving HHVM: the chart shows HPHPc as having flat performance over time, while HHVM was getting better over a 7-month period. Could HPHPc have achieved the same performance gains if the same effort was expended?

I'm not on the HPHP team, but when discussing these results internally, they mentioned that HPHPc performance was improving during this time as well (as you would expect we would need to keep running the site efficiently), and the graph showed point-in-time comparisons of their performance, not from a baseline.

As to whether it would have been possible to get better performance on HPHPc, I'll let people with more knowledge answer.

The benefit of having a single runtime for development and production (and production-like staging environments) is huge, though. There may also be benefits in the push process, which while fast when compared to most other companies, which will continue to deliver benefits for increased speed.

Re: Speeding up PHP with the HipHop VM

#10

Post by maratd » Thu, Nov 29, 2012, 10:34 PM UTC

For those who want to get their hands dirty:

CentOS 6.3 x64: https://github.com/facebook/hiphop-php/wiki/Building-and-ins...

Ubuntu 12.04 x64: https://github.com/facebook/hiphop-php/wiki/Building-and-ins...