From: David Ross <dross@...>
Date: 2004-10-11T11:47:38+09:00
Subject: Re: Performance of Ruby

Remeber now.. you are using windows. You will get a faster statistic if 
you are on a BSD system(doesnt mean linux). The BSD stack is one of the 
fastest implementations. Is there a place to download this benchmark 
program you created?

--dross

Bill Kelly wrote:

>Hi,
>
>From: "Michael Gebhart" <mail@miketech.net>
>  
>
>>I am just thinking about programming a serverapplication with Ruby. The
>>application should manage a lot of sockets and should also be
>>multithreaded. One year ago I have tried to develop a similar application
>>with php. The performance was absolutely bad. No, that was not only my
>>fault ;) My question is now: What about the performance of ruby? Is it
>>possible to write an application using sockets and threads with ruby, but
>>also with a good performance? Normally I am using C/C++ for these things.
>>Sure, Ruby will not be as far as C/C++. But the php-scripts were very very
>>slow. Has anyone experiences with this?
>>    
>>
>
>I've just hacked up a TCP/IP throughput test of sorts,
>where you can spawn a server, and as many clients
>as you want (each client is a separate OS process.)
>
>The server and clients will blast data at each other
>in chunks of a specified byte size, for a specified
>number of seconds.
>
>The server prints some stats about the throughput as
>each client disconnects.
>
>Dunno if it will be useful to you.  One thing of interest
>to me is that the chunk-size makes a huge difference.
>(I.e. printing 2048 bytes at a time is hugely faster than
>printing 64 bytes at a time.  I expected some difference,
>of course, but not as much as I'm seeing.)
>
>With a server transmit chunk-size of 1024, and a client
>transmit chunk-size of 2048... (Arbitrary, just trying
>different values)... Servicing 100 clients, each of which
>connected for 60 seconds, I get:
>
>80.045 seconds, MB in: 190.75 (2.38 MB/sec), MB out: 30.38 (0.38 MB/sec)
>
>I'm running win2k with a 1.33GHz athlon CPU.  The elapsed
>time of 80 seconds is because it took awhile to start up
>all 100 clients (which each ran for 60 sec.)
>
>Leaving the server chunk-size at 1024 but changing the
>client size to 64, I get:
>
>95.728 seconds, MB in: 38.51 (0.40 MB/sec), MB out: 12.15 (0.13 MB/sec)
>
>...And I had trouble even starting 100 client processes
>this time... Well, the same kinds of chunk-size differences
>are apparent when running even one client...   Will have
>to try this on Linux ...  
>
>Anyway for whatever it's worth ... :)
>  
>