> There are _big_ differences between "we have it working" and "we have
> it working and it captures every package on a saturated 100 (1000)
> MBit interface.
>
> mjr clearly stated what they need: zero-copy input BPF. It seems that
> some OS (namely *BSD) delivers this.
Namely, *BSDs do TWO copies, Linux does ONE copy.
His words about zero-copy is apparently a jesture or a provocation.
NFR code (at least its "research" version) is suboptimal and do not
use even capacities provided by *BSD and Linux.
What's about using turbopacket extension with NFR... I do not believe,
that with NFR it is better than common packet socket.
If I did not miss anything, they do not use main BPF advantage of fetching
only part of the packet and grab whole frames. In this case, packet socket
works better than BPF does. And they even need not to hack kernel
(as they do with BSD to increase buffer space :-))
To add, even their blames about Linux-2.0 is in large extent a blatant lie.
In the mode, which they use libpcap, linux-2.0 shows not worse performance
than BPF and adding BPF code to 2.0 is not harder work than replacing
stock BSD BPF with their own hacked version. And the terrible bugs, described
in their FAQ about broken promisc mode etc, either never existed in reality
or absolutely inessential in the worst case.
Alexey Kuznetsov
-
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/