> My pet theory is that their main problem is that they didn't increase the socket
> buffers (64K is simply not enough for 100Mbit/s full speed).
They had to make it for BSD by patching kernel 8)
> The other guess is that they are suffering a bit from the synchronous copy
> (BPF does the copy asynchronously in the kernel while the system is waiting
> for IO etc.).
?? BPF is pretty dumb system copying data twice. The first is copy to buffer
(Linux has not this stage at all), the second is equivalent to Linux way.
> This could be fixed by a libpcap that uses a few worker threads
> for packet recvmsg'ing.
I do not remember a case, when threads helped to get CPU ticks magically.
It really helps to reduce latency a bit by using blocking io instead of
poll/read, but it is inessential when doing bulk work. NFR spends
lots of time analyzing packets, copies lots of useless data,
system overhead is negligible.
> Their main problem probably is that they think Linux is BSD, and it is not @)
Seems, their main problem is brainlessness, unfortunately 8)
Alexey
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/