Re: [IMRG] Requirements for IMP Checkpoint

[James Spooner <jbs3@cs.waikato.ac.nz>]

> There might be another issue to consider here: it might be possible to
extend
> IPMP to switches and other non-IP devices if the TTL match is used only as
a
> trigger.  Once triggered, every device (down to the scale of individual
line
> cards and switches) could (optionally) record information.  Complicated
boxes
> (large routers) could even timestamp more than once if there are multiple
> queues within the box.   This would have to be an optional feature, but if
> you ever talk the likes of Intel and Syskonnect into doing it in standard
NIC
> chips, then it will just happen everywhere....
>
> (The hard part of this would be finding the IPMP header in a device that
isn't
> really doing IP at all).
>

This isn't that hard, it requires buffering a small amount of packet inside
the switch
such that the checksum can be recomputed and spit out the other end.  I have
VHDL
an implementation of IPMP that is instanciated in Xilinx Virtex-II FPGA,
connected
directly to Two Level 1(Intel) 10/100Mbit Ethernet PHYs.  The biggest
problem with
our implmentation was that we had to re-compute the Ethernet Checksum
meaning we
had to buffer around 6 bytes of the packet, 2 for the IPMP pathp and
checksum calculation.

If you would like to see how we did it, I can (with permission from my boss)
make the code
available, our implementation follows the mcgregor draft.

Regards

James


_______________________________________________
IMRG mailing list
IMRG@ietf.org
https://www1.ietf.org/mailman/listinfo/imrg