[bf1942] SMP

[Patrick Gavin]

Besides, you would need access to the source code to get any real 
performance benefits (by using MPI or some other parallelization 
interface). Those benefits may not be tangible due to the latency issues 
that Tom brought up. Most HPC clustering addresses latency with quatrix 
or myrinet, but those solutions usually cost more than the processors 
themselves.

-P

Tom Kiblin wrote:

>Aubrey,
>
>Beowulf clustering, or any clustering won't work well in the gaming 
>world.
>
>We've tried, but the latency between machines is just too great.
>
>Looking at other solutions, as Beowulf just doesn't scale/perform well 
>for game servers.
>
>Tom
>On Sat, 17 
>May 2003 aking at gblx.net wrote:
>
>  
>
>>Just popping in here.. has anyone had experience doing beowulf for bf
>>servers?  This might be a REALLY cool thing to get in on.  Not sure if any
>>other games out there do it or not.  I'm sure clustered bf servers would
>>make bf a solid contender out there, so far as online servers.  Like my
>>buddy, Ricky and I say, 'Who's got that?!  NOBODY!'
>>
>>Aubrey King
>>Global Crossing, Ltd.
>>IP Systems Engineering
>>Systems Administrator
>>
>>On Fri, 16 May 2003, ScratchMonkey wrote:
>>
>>    
>>
>>>--On Friday, May 16, 2003 6:58 PM -0700 Jon <MMmmGood at cox.net> wrote:
>>>
>>>      
>>>
>>>>Well currently we run 5-6 BF servers on dual xeon 2.66s with a max all
>>>>together of about 110 players.  Would be nice though to squeeze a bit more
>>>>out of it though.
>>>>        
>>>>
>>>Multithreading won't do that in that situation, BF is too much of a CPU hog 
>>>so the primary thread will still be tied up for some server. The main 
>>>advantage of multithreading is this case would be to reduce any blocking 
>>>for lengthy stuff, but that can also be accomplished with a suitable state 
>>>machine. The state machine is harder to code and get right, but uses less 
>>>resources than a separate thread since there's no tasking cost.
>>>
>>>      
>>>