In article <7jud0o$q10$[EMAIL PROTECTED]>,
[EMAIL PROTECTED] (Zygo Blaxell) writes:
> In article <[EMAIL PROTECTED]>, Andi Kleen <[EMAIL PROTECTED]> wrote:
>> In article <7jtjfs$jsr$[EMAIL PROTECTED]>,
>> [EMAIL PROTECTED] (Zygo Blaxell) writes:
>> The multipath equal cost routing does not do real load
>> balancing ATM. RTCF_EQUALIZE is not implemented. The -ac kernels contain a
>> small hack to make it work, but it still does not do very well.
>> One Alternative is the teql queueing discipline which works on L2.
> teql does seem to work (although it's hard to tell without rearranging
> the hardware, as all the teql0 traffic is hidden from a tcpdump of the
> slave devices, and tcpdump can't handle teql0 itself). The LEDs on
> both ISP's boxes are flashing in tandem now, so until I get another
> machine in there to tcpdump the net I'll assume it works for now. ;-)
Perhaps the dumb outdated one in RH6 or your dist can't. The PF_PACKET libpcap/tcpdump
should do it fine. Get it from ftp.inr.ac.ru:/ip-routing/lbl*
> It occurs to me that what I really want for balanced receiving is to
> get IP masquerading working with equally selected source interfaces.
> This is orthogonal to whether the individual packets are routed on
> many or only one interface. Since the IP's are different and neither
> ISP knows about routing to the other, the reply packets for outgoing
> connections will always be received on the one interface that matches the
> IP address the outgoing connection was made from. It would be useful if
> the masquerade code could send each entire masqueraded connection through
> each available ISP that could handle the connection. Of course because
> of the port 80 filtering, the list of suitable ISP's is changes depending
> on destination, and would probably have to be controlled by ipchains marks.
Sounds like a job for netfilter.
Easier is to dumb the other ISP though
-A.
-
To unsubscribe from this list: send the line "unsubscribe linux-net" in
the body of a message to [EMAIL PROTECTED]
