Re: [lwip-devel] etharp.c updating

[Leon Woestenberg]

Hello,

On Mon, Jun 13, 2011 at 4:52 PM, Bill Auerbach
<address@hidden> wrote:
>>> In etharp_send_ip(), verify if the matching ARP entry is about to
>>> expire and if so, send out a ARP request right after sending out the
>>> IP packet.
>>
>>Yes, I've used this approach in other stacks and agree that this is the
>>best way to keep the ARP cache valid with minimum overhead.
>
> Although for those of us pushing 300mbS transmission speeds I'm pretty
> confident this will have a noticeable if not substantial impact.  And I have
> not seen the problem that is supposed to occur without this proposed change.
>
Are you using the trusted IP packet for ARP updates option?

What if your ARP entry runs out? You'ld certainly be looking at at
least the ARP latency plus the TCP retransmission latency for a lost
packet (if queueing is off).

Thus worst case, queueing is off, and a TCP packet is lost. I'm
measuring 1 second drop-outs if the ARP cache has to be repruned in
combination with packet loss and TCP retransmit.
I'm not sure why though, I'm seeing 5 calls into tcp_fast_tmr before
TCP retransmits. Maybe there is hold-off involved there or a sub
optimal implementation for that case.

Regards,

-- 
Leon