[e2e] Packet Corruption in a Wireless Medium!

[davide+e2e at cs.cmu.edu]

> Can the headers of ALL packets that are corrupted while traversing
> a wireless link be retrieved by the MAC layer at the receiver
> (as proposed by the Header Checksum Option), so that the ELN
> mechanism can detect packet corruption?

No.  The details depend on both the modulation in use and the
host interface chip.  But in general a sufficiently bad error
burst (or other problem) at the right time will cause the receiver
to fail to sense an incoming packet, or perhaps to consider the
packet to be addressed to another station.  Many systems transmit
this sort of metadata with extra redundancy, but it is still
susceptible to fatal corruption.

Also, if the network interface uses a standard Ethernet controller
chip as a building block, it may be difficult to convince it to
give you *all* of the bad packets.

> If no, what percentage of them can be retrieved successfully?
> Are there any papers that address this issue?

An experiment which I should have done but didn't think of until
too late would have been interesting:  in a high-error environment,
collect low-level error traces (http://www.cs.cmu.edu/~davide/wltraces/)
using multiple receiving stations, and see if the receivers perceive
the same packets as "lost".

Aside from the problem of how many of the bad packets you can collect,
there is the issue of deciding which flow any given bad packet belongs
to.  Unless the destination address, protocol id, and port fields are
well-protected by the physical layer, you could easily generate
loss notifications for the wrong flow.

Dave Eckhardt