On Monday, 16 April 2012 at 07:46:11 UTC, Brad Anderson wrote:
On Sun, Apr 15, 2012 at 6:27 AM, goughy <[email protected]>
wrote:
Hi,
I have just finished the first cut of an HTTP library to
provide
application level support for HTTP (ie as a server, not a
client). The
library provide a very simple mechanism for processing HTTP
requests both
synchronously (via delegates) and asynchronously (via
std.concurrency)
using request/response objects.
eg.
import std.stdio;
import protocol.http;
int main( string[] args )
{
httpServe( "127.0.0.1", 8888,
(req) => req.getResponse().
status( 200 ).
header( "Content-Type", "text/html"
).
content(
"<html><head></head><body>**Processed
ok</body></html>" ) );
return 0;
}
The code is available on github: https://github.com/goughy/d/**
tree/master/http4d
<https://github.com/goughy/d/tree/master/http4d>
Note that this is alpha quality code, so YMMV. I would
appreciate some
feedback, critiques, opinions etc. - particularly in the area
of it being
"idiomatic" as I'm still finding my feet in the D language.
The good thing is without any real performance tuning, I can
push through
over 23,000 requests per second, and I'm sure there is plenty
of room for
improvement.
Cheers.
Neat. I like the integration with std.concurrency and the API
in general
(though I too am still trying to get a feel for idiomatic D).
How does
23,000 requests/sec compare to, say, nginx's performance (which
seems to be
the current performance king)? I just tried to look up some
general
performance figures but they are all over the place and don't
correspond to
your machine in any case.
I just watched the Go talk from Lang.NEXT and he uses Go's
standard HTTP
library and it made me a little jealous. Perhaps this would be
a good
candidate for Phobo's HTTP library.
Regards,
Brad Anderson
I had a look at the following and thought 23k seemed pretty
reasonable for a first cut:
http://timyang.net/programming/c-erlang-java-performance/
But then I also saw
http://nbonvin.wordpress.com/2011/03/14/apache-vs-nginx-vs-varnish-vs-gwan/,
and G-WAN seems to be able to crack 120k, so this library is
unlikely to hit that mark.
But again there's benchmarks and benchmarks, and I haven't tuned
any kernel params etc - just straight out of the box. There is
some pretty naive code also that I'm sure could be better tuned,
but it's currently more than enough for my planned uses, and I'm
happy its in that range without, really, a huge amount of effort.
I would be stoked if this was usable enough for Phobos, but the
dependency on Zeromq might preclude that.
Cheers.
