[Soekris] net4801 throughput with no IPSEC

Soren Kristensen soren at soekris.com
Fri Apr 15 18:16:14 UTC 2005


Hi Michael,

Michael Hamerski wrote:

> Soren Kristensen wrote:
> 
>>
>> Strange, the WRAP and net4801 use exactly same processor and ethernet 
>> controllers, and should be expected to get exactly same performance....
> 
> That would be my expectation too. I'm clueless as to the reason, if any.
> 
>> Is it biased benchmarking, or do is there any BIOS chipset programming 
>> that I missed ?? Do anybody happens to know ?
> 
> I doubt it's inherently biased, might be a freak result somehow. I can 
> try to reproduce the test on a WRAP to ascertain the results.

No, probably not. It could be using the old driver that don't handle 
shared interrupts very well, the FreeBSD driver used to have issues with 
that, but it was fixed.

But I have been told that the benchmarks would be redone using newest 
drivers....

And before anybody asked why I just didn't give the ethernet controllers 
a interrupt each:

The Geode SC1100 have limited ethernet pins available, and since I 
needed more than available individually I decided it to be better that 
devices that would use the same driver shared an interrupt. I was just 
my bad luck that the FreeBSD driver didn't handle that very well and my 
performance testing was with all 3 ethernet ports set up....


Best Regards,


Soren Kristensen





More information about the Soekris-tech mailing list