[Richard Sharpe <rsharpe@xxxxxxxxxx>]

On Mon, 3 Jun 2002, Mark Burton wrote:

> 
> From: Guy Harris <guy@xxxxxxxxxx>
> Subject: Re: [Ethereal-dev] iSCSI dissector and packets with only data ...
> Date: Mon, 3 Jun 2002 13:51:17 -0700
> 
> > On Mon, Jun 03, 2002 at 01:46:02PM -0700, Bill Studenmund wrote:
> > > Another one would be that if there was a preceeding PDU whose TCP write
> > > didn't contain all of the data, then chances are this packet is just data.
> > 
> > Is that "PDU" in the iSCSI sense, or "PDU" in the sense of a TCP
> > segment?
> > 
> > (I.e., is this an issue that doing TCP reassembly for iSCSI - which is
> > supported - would handle, or is there another issue at the iSCSI
> > protocol layer?)
> 
> If TCP reassembly is enabled and no packets are lost and the stream
> content is legal iSCSI speak then the current packet-iscsi.c should
> not mistake data for iSCSI PDUs as long as the first packet it sees
> really is an iSCSI PDU and not data.

Yes, this is correct.
 
> If any packets are dropped and the stream can't be desegmented it all
> goes to worms which is where better heuristics would be useful.

However, we have gigabytes of data going between our managers and our 
OSDs. It is very easy to start recording in the middle of a stream :-(

Regards
-----
Richard Sharpe, rsharpe@xxxxxxxxxx, rsharpe@xxxxxxxxx, 
sharpe@xxxxxxxxxxxx