> > The incorrect checksum value for the first curdled packet turns out to be
> > the CORRECT checksum value for the next curdled packet, whose incorrect
> > checksum value was the correct checksum value for the 3rd curdled packet,
> > whose incorrect checksum value was the correct checksum value for the

> This very probably isn't a bug, it is what happens when you lose a
> packet on a circuit running VJ header compression.  The decompressor
> stays out of sync with the compressor until the far end retransmits,
> which tells the compressor to reset the session state.  See the
> section on error handling in RFC 1144.

Hmm, that makes sense; when I disable VJ compression, I still see pauses
which tcpdump tells me are caused by missing packets.  Where the packets
go is a mystery, because the PPP error count does not increase, nor does
the TCP error count or IP error count!

In the tcpdump without compression, there was an LCP echo response packet
in the gap in the data stream.  I have this horrible nagging suspicion that
pppd managed to read the data packet along with the LCP echo response.