The IP over ATM Mailing List Archive by date

[Werner Almesberger <almesber@lrc.epfl.ch>]

> The AAL5 specification calls for AAL5 to do a CRC for each cell.

Actually, it's only one per PDU ...

> With this being done, and the fact that lines today
> are better than when TCP was first implemented, is there
> still a need to do a checksum in TCP when running over ATM?

"better" is relative ... but indeed, the CRC-32 provides a much better
protection than the TCP checksum.

> I was wondering if most people think not doing the checksum
> would be safe.

I think it is reasonably safe, if:
 - there's no additional segmentation/reassembly ("bridging") between the
   systems that terminate the TCP connection
 - the network hardware in the end systems can be trusted to transfer the
   data reliably to host memory (or whatever the destination in the
   system may be)

> This would save on one of the biggest CPU users in TCP/IP processing.

Yes, checksums are pretty bad on low-end systems, particularly if they
prevent you from using single-copy (see also my paper
ftp://lrcftp.epfl.ch/pub/linux/atm/papers/atm_on_lowend.ps.gz)

> Any comments or reasons why this should not be done?

I'm afraid that somebody will eventually come up with some "clever"
device playing "ATM network" at both ends and secretly re-computing
the AAL5 CRC in the middle.

- Werner

-- 
  _________________________________________________________________________
 / Werner Almesberger, DI-LRC,EPFL,CH   werner.almesberger@lrc.di.epfl.ch /
/_IN_R_311__Tel_+41_21_693_6621__Fax_+41_21_693_6610_____________________/