Re: ISCSI: Urgent Flag requirement violates TCP.

[Matt Wakeley <matt_wakeley@agilent.com>]

Glen Turner wrote:

> Matt Wakeley wrote:
>
> > > But that's an argument for iSCSI *not* using TCP services beyond the
> > > socket API.  To use such services the general purpose
> > > computer running TCP/IP offload would then need to change the
> > > code in the operating system *and* the code in the ethernet adapter.
> >
> > Yeah, what's wrong with that? It sounds to me like you're saying "we
> > can't make advances in speed and technology because we have to
> > change the OS".
>
> I am saying that the WG needs to be more upfront about its proposal
> to alter TCP.

Nowhere does it say that TCP is going to be altered!

Those that want better performance will alter their implementation of it (move
it out of the OS, move it into hardware, etc).  In doing so, they must provide
the exact same interfaces to TCP as if it resided in the OS.  There is no "new
service" as you describe below.  Why do you think RFC793 needs to be changed
just because something is moved from a software implementation to a hardware
implementation?


> To "make an advance in speed and technology" by altering the services
> offered by TCP, a draft that updates RFC793 should be issued.  The
> services are listed at the start of RFC793, before the description
> of the protocol that implements them.
>
> If the draft becomes a standards-track RFC then operating system
> authors can update their TCP implementations to provide the
> new service.
>
> The current iSCSI proposal requires the use of a non-existent TCP
> service -- that is, the delivery to the application of indexed
> octets and the delivery of indexed urgent data.

Where does it say anything in the iSCSI proposal about indexing?  If you're
thinking about the delivery of the "urgent byte" to demark the beginning of
the frame, that is already provided for in existing implementations of TCP.
From the getsockopt man page:

      If SO_OOBINLINE is set, out-of-band data (TCP "urgent data") is left
      "in-line" among the normal data stream.  In that case, the SIOCATMARK
      ioctl() request must be used to determine if the inbound data stream
      has been read up to the point where the out-of-band data begins.  If
      multiple transmissions of out-of-band data are received before the
      application reads them, all of the data is left in-line

>  RFC793 is plain
> that the current service is not indexed, but is a stream of octets.
>
> Commercially, my belief is that if iSCSI uses a new TCP service

iSCSI is not using anything new.  Please point out what service you think is
new.

>
> then the protocol is constrained to being a cheaper alternative
> to Fiber Channel for enterprise storage.  Whereas the use of
> existing TCP/IP services allows iSCSI to be used to implement
> "global storage" (to create some market-speak).
>
> As you may have gathered, it is the global storage potential
> of iSCSI that has attracted me to this WG.  Enterprises
> sharing high-value storage assets, ISPs offering file-system
> independent backup facilities, etc.
>
> Best wishes,
> Glen
>
> --
>  Glen Turner                                 Network Engineer
>  (08) 8303 3936      Australian Academic and Research Network
>  glen.turner@aarnet.edu.au          http://www.aarnet.edu.au/
> --
>  The revolution will not be televised, it will be digitised

-Matt Wakeley
Agilent Technologies