Importance of DDP checksums?

[Peter Poorman]

Can someone tell me why (and whether) DDP checksums are important?

I've observed empirically that Macs generally don't generate one, but some
routers do. I've also noted that at least one printer seems to ignore all
packets that have non-zero checksums.

Inside Appletalk wasn't much help (just says that they help insure integrity
at the datagram level). The MacWorld Networking handbook didn't shed much
light either.

Some explaination in end-user terms would be greatly appreciated.

Thanks!

-- Pete Poorman
poo...@convex.com

[David Stine]

In article <poorman....@convex.convex.com> poo...@convex.com (Peter Poorman) writes:
>Can someone tell me why (and whether) DDP checksums are important?
>

Well, the short answer is "yes."

The longer answer is "yes, but they would be much more useful if (a) all
AppleTalk nodes generated DDP packets with checksums and (b) all AppleTalk
nodes would generate DDP packets with the _correct_ DDP checksum."

>I've observed empirically that Macs generally don't generate one, but some
>routers do. I've also noted that at least one printer seems to ignore all
>packets that have non-zero checksums.

The problem here is this: The AppleTalk DDP calls let the application specify
whether or not they want the checksum generated on the packets. So some
applications generate and use DDP checksums and others don't.

>Inside Appletalk wasn't much help (just says that they help insure integrity
>at the datagram level). The MacWorld Networking handbook didn't shed much
>light either.

Generally speaking, if you want highly detailed information about what/how the
AppleTalk drivers do things on the Macintosh, you first read the documentation
and then reach for your MacNosy manual and start disassembling the code.

The DDP checksum would be useful for finding memory or other data-corruption
errors in intermediate routers and bridges if it were used consistently.
Since it is not generated or checked consistently across an AppleTalk network,
my advice to you is that if you really want to insure end-to-end data
integrity, you should generate your checksums at the application level and
check it on the receiving end.

dsa

[Jim Matthews]

Peter Poorman writes

> Can someone tell me why (and whether) DDP checksums are important?

The legalistic reason is that they are defined in the specification of the
protocol, so any implementation that does not handle them correctly (such
as the original Personal LaserWriter NTR ROMs, the AppleTalk ImageWriter
ROMs, etc.) is *broken*.

The practical reason is that while link-level protocols (such as ethernet)
typically have their own checksums, packets can be corrupted while moving
from the network interface to the computer/router's memory, or while in
the computer/router's memory. DDP checksums are a way to keep these
corrupted packets from being forwarded and accepted; failed checksums are
also an early warning sign of hardware problems in routers and interface
cards.

At Dartmouth we checksum everything, and it's one reason why we've been
able to maintain a 6,000 node AppleTalk internet with just a couple
network technicians. The biggest problem we have with checksums is that
some implementors handle them incorrectly, and then we have trouble using
their broken products. If more sites generated and checked checksums it
would be much less likely for these products to get out of beta test
without being fixed.

Jim Matthews
Dartmouth Software Development