Hacker News Reader

Comment by 1SaltwaterC | Hacker News Reader

There's no "G-WAN's Javascript engine". Just plain old V8 embedded via a handler. You get just the basics, no batteries included aka you need to code the bindings to various things (like FS, crypto, etc) by yourself.

A year ago, a member of the former G-WAN community wrote a handler and V8 wrapper: https://gist.github.com/1077769. Guess that the G-WAN author (Pierre) tweaked that a little bit more.

Tested it myself. For simple "hello world" it was 5.5x faster than node. Quite impressive given the fact that a new V8 context is created with every request. But, again, no batteries included.

View on HN

That makes me all the more curious: why is it faster? I cannot imagine Node doing something obviously wrong that would explain the difference. Evented I/O is pretty well understood, I/O loops aren't that hard and it's all just accept()/read()/write()/close() at the end of the day. Where does the 5x difference come from?

It is plausible though. node.js have its fair share of "slowness" when compared to other async frameworks for the basic "Hello World" (shameless plug): http://www.saltwaterc.eu/async-frameworks-hello-world-showdo...

vert.x used the JavaScript bindings via Mozilla Rhino. I guess that a pure Java test could be a little bit faster than that.

Because it caches (http://gwan.com/faq#cache) response under high concurrencies. Try to benchmark this code:

  #include "gwan.h" // G-WAN exported functions
  int count=0;
  int main(int argc, char *argv[])
  {
   if (!count) printf("New Counter!\n");
   xbuf_xcat(get_reply(argv), "Hello World %d",count++);
   return 200; // return an HTTP code (200:'OK')
  }

Wow, I didn't know that. I certainly did not expect to find that information in the FAQ. In fact this is the first time I've heard of a web server doing that.

So was the benchmark the author posted with microcaching turned on? While it's a legit way to reduce the load in real-world situations, it feels a little like cheating when used in a benchmark.

Reply on news.ycombinator.com