[Beowulf] gamers: evil or just useless? ;)

Douglas Eadline deadline at clustermonkey.net
Mon Aug 14 06:36:15 PDT 2006


> On Tue, Aug 08, 2006 at 06:41:14PM -0400, Mark Hahn wrote:
>
>> first GPUs, now NICs?  http://www.bigfootnetworks.com/ appears to be
>> an actual linux-based coprocessor marketed as an offload NIC for gamers.
>> weird, but true!  from their whitepaper, I can't tell whether their
>> performance numbers are sane (2500 UDP "calls" per second sounds very
>> underwhelming...)  will game vendors actually write code specific
>
> Their white paper's little bar graph claims that its "UDP throughput
> in megabytes" is "20.15", while that of an Intel Pro 1000 card is only
> "11.77".  They make no attempt to explain the origin of that delta.

I have no idea what this card does, but other data points are
always useful:

Linux kernel 2.6.14, Intel Pro 1000 desktop (32 bit/33 MHz)
1500 byte MTU, netperf gives 66 MBytes a second using a
message size of 1472 (44,701 messages per second).
The hardware was two 1.7 Ghz Sempron systems.
BTW, netpipe TCP latency was 32 us (interrupt throttle
rate off - this may not be the best setting
for every code, however)

 --
 Doug




>
> Assuming those numbers are in fact correctly measuring something
> relevant, perhaps their card just has its network settings tuned
> correctly for that test, while the Pro/1000 does not?  Or the OS
> networking stack on the desktop in question is lousy, which degrades
> the Pro/1000 but not their card, as their card is doing OS-bypass?
>
> (They specify the motherboard, etc. of their test system, but don't
> even mention what OS it was running.  MS Windows presumably, but they
> say nothing about what version.)
>
> I can imagine three sorts of scenarios for what this card is really
> doing:
>
> 1. It might effectively be just a plug in key to change your Windows
> network settings.  In which case, well, $280 seems a little steep.
>
> 2. It might basically be a way to sell bundled OS-bypass networking
> software for MS Windows.  I suppose this might actually be cost
> effective for some people.
>
> 3. It might actually be beter / more interesting / more useful than
> any of those "Let's take advantage of naive gamers!" scenarios above.
>
> I wonder which is closest to the truth.  #3 of course would be the
> most interesting, but I don't think they've provided enough data to
> even distinguish between cases 1, 2, and 3, much less understand #3.
>
> --
> Andrew Piskorski <atp at piskorski.com>
> http://www.piskorski.com/
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit
> http://www.beowulf.org/mailman/listinfo/beowulf
>






More information about the Beowulf mailing list