Re: Avoiding deadlock in iSCSI

[meth@il.ibm.com]

The current iSCSI draft allows for (successful) status to be sent with the
last data PDU. This should also be done in the asymetric case, precisely
for the reason articulated by Pierre. The NIC receiveing the data will know
when to perform the interrupt when the data transfer has completed and
return a good status, and only one interrupt will be required to complete
the data transfer. (The data buffers are registered to the NIC together
with the Initiator Task Tag [or Transfer Tag], so successful status
received on the data NIC can be easily associated with the original
command, and there is no need for an interrupt on the command NIC.)

- Kalman

Pierre Labat <pierre_labat@hp.com> on 13/09/2000 18:57:57

Please respond to Pierre Labat <pierre_labat@hp.com>

To:   ips@ece.cmu.edu
cc:    (bcc: Kalman Meth/Haifa/IBM)
Subject:  Re: Avoiding deadlock in iSCSI





<... deleted ...>

The fact that in the Kalman draft the end of data notification and
command status
comes on two different TCP connection (thus 2 NICs most of the time)
implies
an important
CPU overhead on the initiator to manage the READ command completion as
described
by Somesh Gupta. You need for each READ to work accross two NICs. You will
double
the work on each read completion. Moreover you will double the number of
cache misses
because you need to read completion information from two NICs. And cache
misses
are really stalling your CPUs.
It's a reason for what i can't agree on the Kalman proposal. You don't have
this penalty
in the proposal below.

Regards,

Pierre


,... deleted ...>