Re: iSCSI: Multiple markers before end of a data PDU

[Stephen Bailey <steph@cs.uchicago.edu>]

Barry,

> If markers are being used a there is a long PDU with a short marker interval
> such that multiple markers come before the end of the PDU, are the bytes (as
> measured on the wire) counted in the offset to the start of the PDU that is
> the value of the marker.

Yes.

In other words, the TCP sequence number of the start of the PDU
pointed to by a marker is TCP sequence number of the byte following
the marker, plus the `Next PDU Offset' value contained in the marker.

This makes markers a bit more complicated to insert, but easier to
process on the receiver (which is working harder in other ways
already).

> If the marker value is to be the TCP byte stream offset as seen on the wire
> then the process that does the insertion of the markers will need to know
> the size of the PDU, the position of the last marker and the marker
> interval.

Yes.

Note that it probably doesn't make a lot of sense to have a short
marker interval and a long maximum PDU size.  The minimum grain size
to which you could possibly bound your eddy buffer is O(PDU size).
You just need to have markers frequently enough that you'll be able to
find the beginning of the next PDU with high probability.

In that sense, a maximum PDU which is roughly MSS bytes and markers
which occur roughly every MSS bytes gives you about the best bound you
could hope for.

Steph