RE: iSCSI: IP fragmentation

[Dennis Young <dyoung@rhapsodynetworks.com>]

Josh,
I think this is a good idea.
Does this mean we should do a MTU discovery on the path between the 
iSCSI initiator and target so that we can do intelligent DataPDULength 
selection.
Maybe the default is to keep the DataPDULength at 2 units (1 unit = 512
bytes)
to avoid fragmentation, assuming the interface at the two ends are Ethernet.
-dennis

-----Original Message-----
From: Joshua Tseng [mailto:jtseng@NishanSystems.com]
Sent: Thursday, June 07, 2001 6:10 PM
To: ips@ece.cmu.edu
Subject: iSCSI: IP fragmentation


		Folks,

		IP fragmentation is an issue which will affect the
performance, logic and
		HBA/NIC memory requirements of iSCSI implementations. Note
that
		fragmentation by intermediary switches and routers is not
permitted
		in IPv6 (hence, the DO NOT FRAGMENT bit does not exist in
IPv6).
		With this in mind, and in the interest of optimizing iSCSI
performance,
		I was wondering if it would make sense to require that the
		DO NOT FRAGMENT bit to be set on IPv4 packets carrying iSCSI
and
		that iSCSI implementations not perform IP fragmentation for
IPv6.
		This will allow for performance optimization if
implementations do not
		have to check for fragmentation and handle reassembly.

		Any thoughts?

		Josh