RE: [lwip-users] Jumbo frame questions

[Bill Auerbach]

> Kieran Mansley
> Sent: Tuesday, February 05, 2008 10:24 AM
> To: Mailing list for lwIP users
> Subject: RE: [lwip-users] Jumbo frame questions
> 
> There may be a bug here, but I think it's just you're hoping the lwIP
> has a feature that it doesn't have: path MTU discovery.  Normally if you
> try and send a packet with a larger MTU than some piece of the network
> can't handle it will either fragment it (if it's allowed) or drop it and
> send an ICMP message back to the sender.  I expect that lwIP does
> nothing with this ICMP message.  This could be construed a bug
> (especially if we also set the DF - Don't Fragment - bit in the header)
> but could also be seen as a property of lwIP being "lightweight".
> 
> For the avoidance of confusion:
>  MSS = largest TCP payload
>  MTU = largest IP packet (i.e. MSS + IP and TCP headers).
>  Frame size = largest ethernet frame (i.e. MTU + ethernet headers etc)
> i.e. You may need to adjust your MSS settings as it looks like you're
> giving it the values you want for the MTU or frame size.

This may be a big part of my problem.  I was trying to configure off of one
#define to build either Jumbo or normal lwIP.

If I have

#define PKTSIZE 1514

And

netif-mtu = PKTSIZE - 14;

And

TCP_MSS (PKTSIZE - 54)

This gets me an MTU of 1500 and MSS of 1460.  This works much better now.

This works now with the PC's Ethernet controller at 1514 or 9000.  I know
you say it's not supposed to, but I do see frames over 1514 in WireShark and
I can still access our LAN and WAN.

Thank you for setting me straight on these 3 values.

What are the reason(s) you wouldn't set MTU and MSS based on the packet size
as I have done?

Bill