Re: aws vs ruby rails or php? how much faster?

["Randy Brukardt" <randy@rrsoftware.com>]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 2032 bytes --]

"Yannick Duch�ne (Hibou57)" <yannick_duchene@yahoo.fr> wrote in message 
news:op.wwhg52xpule2fv@cardamome...
Le Thu, 02 May 2013 23:13:24 +0200, Randy Brukardt <randy@rrsoftware.com>
a �crit:
>> A bit of background: Janus/Ada uses it's own task management, so all 
>> tasks
>> run inside of a single OS process/thread. In our web server running on
>> Windows 2K, we tried to use asynchonrous socket I/O to prevent unintended
>> blocking from socket calls. But the effect was that the server ran 10 
>> times
>> slower than using busy-waited socket I/O; effectively W2K used only a 
>> single
>> OS thread to implement then asynchronous sockets and that effectively
>> serialized everything going in and out of the server.
>
>I don't understand how this can make it working ten times slower. W2K must 
>be eating something in the while, which is not related to the asynchronous 
>I/O. Or else, I don't understand.

It's as I explained; the "asynchronous sockets" were apparently implemented 
by a single Windows thread. So if multiple I/O were done more or less at the 
same time (which is very common on a web server that's receiving requests 
and sending replies (files)), Windows did each one of them in order -- the 
second one had to wait for the first to finish before it was even started. 
The net effect was to turn the entire server into a single-threaded program, 
even though the server itself never blocked.

The speed effect seemed like a factor of ten or so; with the supposedly 
asynchronous sockets, it could only handle one or two simultaneous requests. 
The busy-waited version can handle up to 20 requests at a time, although the 
size of our DSL connection tends to limit the number of successful 
connections to 8-12 before everything slows to a crawl.

It will be interesting to see how this behaves on Debian using GNAT. (This 
is essentially the worst case for porting Ada code; different compiler on a 
different target with different bit sizes.)

                                             Randy.