Re: ISCSI: Urgent pointer consensus

[Matt Wakeley <matt_wakeley@agilent.com>]

Silvano Gai wrote:

> >The benefit to analyzers was only a side benefit.  The main reason is to place
> >data in memory where it belongs in the presence of lost frames.  See my note
> >"ISCSI: Why the use of the Urgent Pointer"
>
> You can do that by writing the payload of TCP packets in memory at an offset given by the "Sequence Number". This will leave holes in the memory for missing packets. You will fill these holes, when packets are received correctly. No need to copy or move a single byte.

No you can't.  The way SCSI works is the host provides pre-assigned buffers to the driver where it expects the data for a particular I/O to be placed.  If you lose an iSCSI PDU header due to a lost TCP segment, you lose iSCSI PDU framing from then on (until the missing
segment is received).  You then have to store the TCP data (received after the missing segment) in a generic holding pool and wait until the missing segments arrive.  Remember, you need to examine the iSCSI message header to determine what the message is and to what I/O
it belongs to. When the missing segments arrive, an adapter will then have to move this "piled up" data off the adapter to the host, as well as move the new incoming data to the host as well.  This will put an extreme strain on existing backplanes, especially at 1Gbps and
higher.

With an iSCSI framing mechanism, an adapter will only have to buffer data up to the next PDU header (marked by the urgent pointer).  It can then continue processing iSCSI message PDUs and placing the data into the appropriate pre-allocated host buffers.

Matt Wakeley
Agilent Technologies

>
>
> -- Silvano
>
> Silvano Gai
> Cisco Systems                  Empowering the Internet Generation
> 170 West Tasman Drive                         Tel. (408) 527 2690
> San Jose, CA, 95134-1706                      Fax. (408) 527 2690
>                                                    sgai@cisco.com