RE: Re: multiple tcp connections

[psarkar@almaden.ibm.com]

I would second that, we have also managed to obtain link speed
on a single connection,


   Prasenjit Sarkar
   Research Staff Member
   IBM Almaden Research
   San Jose


somesh_gupta@hp.com@ece.cmu.edu on 07/10/2000 11:04:54 AM

Sent by:  owner-ips@ece.cmu.edu


To:   ips@ece.cmu.edu, Julian Satran/Haifa/IBM@IBMIL
cc:
Subject:  RE: Re: multiple tcp connections



I don't think we want to invent sessions because some implementation
could not achieve link speed using a single connection.

First of all, we have demonstrated link speed on high speed links
with a single TCP connection. There are two reasons I can come up
with on short notice that may prevent one from acheiving this number.

One is that the TCP window size may not be large enough. With the
large window size option, this should not be an issue. The other
is poor design - and that is something that should be solved
through better design.

There may be other reasons why this is not achievable that I
don't know.

--

When we get to adding all the other objectives, that is another
can of worms. We should at least spell out the requirements we
are placing on a session in the spec.

Somesh


-----Original Message-----
From: julian_satran@il.ibm.com [mailto:julian_satran@il.ibm.com]
Sent: Saturday, July 08, 2000 10:38 PM
To: ips@ece.cmu.edu
Subject: FW: Re: multiple tcp connections




John,

Correct - the only difference would be that the higher levels of software
would have to be aware
and handle it (error recovery, exceptions, balancing, increasing the
number, decreasing the number).
Those where the considerations that lead me to invent the session (channel
group in our own slang) -
as we think we can handle those well under the hood.

Julo


hufferd@us.ibm.com on 08/07/2000 21:34:22

Please respond to hufferd@us.ibm.com

To:   julian_satran%ibmil.RSCS@DEVM.DE.IBM.COM
cc:   ips@ece.cmu.edu (bcc: Julian Satran/Haifa/IBM)
Subject:  Re: multiple tcp connections






Julian,
Since a session is only an iSCSI thing, your statements would be correct
even if there were only one TCP connection per session, but several
sessions on the same NIC.




.
.
.
John L. Hufferd
Senior Technical Staff Member (STSM)
IBM/SSD San Jose Ca
(408) 256-0403, Tie: 276-0403
Internet address: hufferd@us.ibm.com
Notes address: John Hufferd/San Jose/IBM @ IBMUS
VM address: hufferd at IBMUSM54


Julian Satran/Haifa/IBM@IBMIL@ece.cmu.edu on 07/07/2000 11:18:48 PM

Sent by:  owner-ips@ece.cmu.edu


To:   ips@ece.cmu.edu
cc:
Subject:  Re: multiple tcp connections





Parallelism on end-points could be another reasons.

A long time ago when running some benchmarks we achieved considerably
better numbers with 4 sockets than with one.

This type of result repeated itself in different contexts and with
different stacks.

That is probably due to some serialization that is inherent in the way
stacks are built
and API's are used (this is a long discussion subject).

With SMPs the differences are even more striking.

Julo

Matt Wakeley <matt_wakeley@agilent.com> on 08/07/2000 02:11:57

Please respond to Matt Wakeley <matt_wakeley@agilent.com>

To:   Randy Haagens <Randy_Haagens@hp.com>, Julian Satran/Haifa/IBM@IBMIL
cc:
Subject:  multiple tcp connections




Hi,

I was asked the question:

If there is only 1 MAC (available to the system), would there ever be more
than 1 tcp connection per iSCSI session.  I couldn't think of a good reason
why there would be, but thought I'd ask you. The only reason I could think
of
using multiple tcp connections per iSCSI session was for port aggregation
when
there are multiple MACs available.  Any comments?

Thanks,

-Matt