HHVM 3.10 Brings More Performance Improvements, Improved Reflection

[phoronix]

Phoronix: HHVM 3.10 Brings More Performance Improvements, Improved Reflection

HHVM 3.10 was released today as the newest version of Facebook's interpreter for PHP and their Hack programming languages...

http://www.phoronix.com/scan.php?page=news_item&px=HHVM-3.10-Released

[Michael_S]

I realize Facebook has millions of lines of production PHP code, but I'm astonished that their solution to PHP performance problems was better PHP runtimes and a PHP++ language.

It would seem intuitive to me that on the scale they're operating, it would make sense to put entire teams on reimplementing their entire universe of PHP code in a language that lends itself to better performance. "We have a hundred people working on the PHP code and we keep hitting performance issues." "Put twenty people on a project improve the PHP interpreter. Also, put fifty people on rewriting our PHP code in Haskell and fifty people on rewriting our PHP code in C++. Keep the two rewrite teams separate, I want to see what they come up with."

[wagaf]

I realize Facebook has millions of lines of production PHP code, but I'm astonished that their solution to PHP performance problems was better PHP runtimes and a PHP++ language.

Facebook employs many PHP programmers.

PHP programmers are much easier to find and much cheaper to pay than C++ or Haskell programmers.

[Xelix]

I realize Facebook has millions of lines of production PHP code, but I'm astonished that their solution to PHP performance problems was better PHP runtimes and a PHP++ language.

It would seem intuitive to me that on the scale they're operating, it would make sense to put entire teams on reimplementing their entire universe of PHP code in a language that lends itself to better performance. "We have a hundred people working on the PHP code and we keep hitting performance issues." "Put twenty people on a project improve the PHP interpreter. Also, put fifty people on rewriting our PHP code in Haskell and fifty people on rewriting our PHP code in C++. Keep the two rewrite teams separate, I want to see what they come up with."

This would require changing their current code base AND programmers. On the other hand, if they improve PHP (both the runtime and language), they can keep both their current code base and PHP employees.

[uid313]

I wonder how much of Facebook's code is in PHP and how much is in Hack, and how much is in other languages.

I also wonder what libraries and frameworks they use if any.

It is too bad PHP does not support decorators.

[Michael_S]

Facebook employs many PHP programmers.

PHP programmers are much easier to find and much cheaper to pay than C++ or Haskell programmers.

Sure, but look at the potential return on investment.
1. In theory, a well written C++11 or C++14 program will have fewer bugs than an equivalent PHP program because of the C++ type system. Haskell will almost certainly have fewer bugs due to the Haskell type system.
2. Facebook runs tens of thousands of servers. Now admittedly, much of their performance headaches may be related to scaling storage and networking throughput, and may not be bound by CPU or RAM. But even if more efficient code can't let them reduce their server farm size because of networking or storage needs, just in terms of just CPU usage better code might save them many millions of dollars in electricity.

Again, I don't hate PHP. It enabled Mark Zuckerberg to launch the site, so clearly it gets the job done. But from what I understand, even with the latest improvements Hack is routinely one third as CPU efficient and memory efficient as Haskell. And Haskell, as good as it is, can't run with C++ through GCC or LLVM yet.