ORe + L4Linux network performance
zaytsev at altell.ru
Tue Apr 28 14:34:42 CEST 2009
Björn Döbel wrote:
>> Could anyone please tell me?
>> 1. Did someone measure a L4Linux network performance when ORe is in use? Are
>> there any report or paper with results?
> No reports, sorry. We did run singe L4Linux instances with ORe on some
> 100 MBit ethernet cards (that was about 4 years ago) and got very low
> drops (<5% if at all) in throughput.
>> 2. What about 1 Gbit/s cards (e.g. Intel e1000)? Were they tested?
> We tested ORe on e1000 as well as a TG3 NIC built into one of our test
> machines. Cannot give you performance numbers for that, though.
>> 3. How badly is influence of other running L4Linux processes on network
> I don't assume there will be any difference to the influence normal
> Linux processes have on the network performance. You will get
> performance hits if you run several instances of the L4Linux kernel and
> ORe needs to multiplex the physical NIC amongst them, but this is
> probably expected.
Hallo, Bjorn. Thank you for the swift reply.
We too ran some simple benchmarks, and with bulk tcp transfers
we are getting up to 45 megabytes per second, which is not bad
at all. But it seems that running other processes in Linux does
interfere with ORe. Running a single cpu hog in Linux brings any
traffic to a halt. This is probably due to the Linux kernel and
processes having the default static priority (0x40) higher than
ORe (0x10). Setting the Linux priority to 0x09 fixes the problem.
So, is there a reason to have the default Linux priority that
hight, or was it just a mistake?
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.
More information about the l4-hackers