Re: Assymetric connections [WAS: Re: iSCSI Autosense Consensus, Connection next steps]

["Randall R. Stewart" <randall@stewart.chicago.il.us>]

csapuntz@cisco.com wrote:
> 
> Hi Randall,
> 
> > > * iSCSI data/ready to transmit (rtt)
> > >         - potentially high bandwidth
> > >         - with appropriate headers, segments can be
> > >           processed entirely out-of-order
> > >         - simple data transfer state machine
> > >         - no flow control on data needed, just congestion control
> > >                    - rtt's need to be flow-controlled, but
> > >                      that can be done with a simple credit mechanism
> > >         - data needs to end up in special buffers (e.g. buffer cache)
> > >
> > I understand how you will obtain all of the above with
> > two TCP connecions. I can even support this, considering
> > the 1 for control is "low bandwidth" (note: I can't support
> > N data connections.. only 1).
> 
> Thanks for pointing this out. The wording was a bit loose.
> 
> How does this sound?
> 
> Multiple parallel data connections should only be used in the
> presence of an ECM-like congestion layer that does congestion
> control on macro-flows.

Much better :) but are you going to use the standard SHOULD/MUST/MUST
NOT
conventions? I did not see the standard pointers in the document
and I also did not see any MUST/SHOULD's in here..

I think the above needs to not be a should but more a MUST... how about

"
If multiple parallel data connections are to be used, they MUST
be used with an ECM-like congestion layer that does congestion
control on macro-flows.
"

Also you may need to put some references in to ECM or describe
it some where :)

> 
> > But my question is to the Data connection. How do you get
> > "no flow control on data needed, just congestion control" on
> > a TCP connection?
> 
> You don't, but you can approximate it by opening the window to 1-2
> gigabytes.
> 

As long as the data buffers are there to support this it should
be no problem. I have found in my playing that often times if
the kernel underneath has some buffer set (say 64k in a solaris
TLI UDP stack) a user application above this that tries to set a
window above 64k is wasting its time i.e. you start getting data
losses at the host, a very sad occurance when you transport it
all the way across a large IP network :0

But I have to assume we are talking about not your ordinary TCP/IP
stack here by the things that are being asked... or one that has
some extra knobs to turn...

I am still unsure how you will get unordered messages across though...


-- 
Randall R. Stewart
randall@stewart.chicago.il.us or rrs@cisco.com
815-342-5222 (cell) 815-477-2127 (work)