Cell Relay Archive

[georgem@nntp.best.com (George Marshall)]

Mr M.A. Simmons (m.simmons@ee.surrey.ac.uk) wrote:
: Can anyone confirm/correct our thoughts on error detection with native
: ATM and AAL5:-

: If errors occur in ATM cells then the CRC component at the end of the
: PDU will no longer be correct on the receiving station.

: If this is the case then AAL5 will discard all cells comprising the 
: CPCS PDU (up to about 64k).

: Therefore, it would be up to the higher layers to organise a request to
: retransmit the information.

Must be a hot topic at Surrey! - Two posts on the same subject,
back-to-back...

OK, here's the deal.

The ATM layer defines how to multiplex different
streams/connections/whatever over the same physical link, at a
granularity of 48 bytes.  The various AAL's (ATM Adaptation Layers)
above ATM define how various types of information, from circuits to
packets, can be carried in a standard way over ATM (so far, we have
three: AAL1, AAL3/4, and AAL5 - AAL2 is the eternal placeholder, never
completed, but often quoted).  AAL5 is one of those protocols; it 
defines how to carry packets over ATM in a rather efficient way, but 
AAL5's scope is limited to how to delineate a packet over a virtual 
circuit on ATM.

In particular, as pointed out, AAL5 is not a reliable transport protocol -
if you lose a cell, you lose the packet it's in (just like TDM networks -
losing a bit probably equals losing the packet it's in).  There are multiple
choices for providing end-to-end reliability - if we're in the internet
world, then TCP is presumably providing retransmission services.  In a
"pure" ATM application, if reliable transport is needed, then SSCOP or an
application-specific protocol would be used.

Regards,

George Marshall