[prev in list] [next in list] [prev in thread] [next in thread] 

List:       openbsd-tech
Subject:    Re: jumbo frame support for sk(4)
From:       Ryan McBride <mcbride () openbsd ! org>
Date:       2004-07-31 21:53:45
Message-ID: 20040731215345.GA3756 () countersiege ! com
[Download RAW message or body]

On Sat, Jul 31, 2004 at 08:25:52PM +0000, Ryan McBride wrote:
> - Set size correctly for TX dma map.
> - Implement our own external mbuf storage manangement (based on the
>   jumbo frame support in if_bge.c). There is no separate rx ring for
>   jumbo frames, so we use the external mbuf storage for _all_ packets.
> 
> This diff is a major overhaul of how the sk driver handles incoming
> packets, so it needs solid testing in non-jumbo environments as well.
> 
> If you'd like to help me out even more by testing the jumbo support, the
> em(4) and ti(4) Gigabit Ethernet cards also have working jumbo frames.

Testing jumbo support can be as easy as pinging between the two
machines, or it can be more involved, like testing NFS.

To enable the use of jumbo frames, the mtu must be increased on all the
machines you're testing between:

# ifconfig <if> mtu 9000

Then you can simply try pinging with large packets:

# ping -s 8000 <other-machines-ip>

The application that can most benefit from jumbo frames is NFS, so
testing that would also be helpful: nfs sends 8k packets which are
currently fragmented; an mtu of 9000 allows those packets to travel
intact and reduces the number of interrupts as well as avoids the cost
of IP fragmentation and reassembly.

-Ryan

[prev in list] [next in list] [prev in thread] [next in thread]