RE: FCIP: A question about framing

["Bernard Aboba" <aboba@internaut.com>]

>> Might I suggest that a more profitable approach would be to
>> spend some time coming up with set of requirements for the
>> transport?

>Well, I am not saying that this is a bad idea :), but
>the sigtran working group as just got through doing
>just that... and defining a transport as well. It
>took close to two years by the way :)

The reason that I think that a transport requirements
document would be helpful is that there 
are really two issues that are being addressed 
here at the same time and their interaction complicates
the discussion. 

One issue is the need for extremely high speed. 
With 10 Gbps, 100 Gbps, even 1000 Gbps 
technology coming down the pike in the next 
decade, we are going to need new architectures 
in order to keep up.  

Understanding the impacts and 
needs is probably a major task by itself.  
iSCSI may be the first application
to tackle this issue, but it will probably
not be the last. Thus, I would hope that this
issue is handled with some degree of generality. 

The second issue is the specific SCSI and FC over
IP transport and security requirements. In 
discussing this, it is easy to make implicit 
assumptions about the likely solutions to 
issue #1, without necessarily laying this out 
explicitly. This makes it hard to tease the 
two problems apart and understand where 
the assumptions and proposals differ and 
what their architectural implications are. 

>Now in looking at SCSI and FC over IP and following
>this list for quite sometime their requirements match
>up pretty closely (as near as I can tell) with what
>sigtran came up with. 

It'll be hard to know for sure until those requirements
are written up in detail. 

>Instead they want to continue to try to either TWIST 
>TCP and make some fundamental changes to it

I think that we recognize that fundamental changes
are needed to transport implementations to operate 
at these speeds (issue #1). However, this does not
imply changes to the transport protocols. The key
to making progress is to clearly differentiate what
issues can be solved within a high speed 
implementation, and what issues are inherent to 
the transport.