W3C home > Mailing lists > Public > xml-dist-app@w3.org > November 2003

Re: Proposal for multi-reference support in MTOM

From: <noah_mendelsohn@us.ibm.com>
Date: Mon, 17 Nov 2003 21:07:30 -0500
To: Marc Hadley <Marc.Hadley@Sun.COM>
Cc: "Xml-Dist-App@W3. Org" <xml-dist-app@w3.org>
Message-ID: <OFDD44FD7D.4359ACFC-ON85256DE2.000BA6A1@lotus.com>

Marc Hadley writes:

> To allay your first concern, I'd be open to
> recommending that the binding check for differences
> during serialization. Computing a low cost hash during
> serialization wouldn't add much overhead I suspect.

Maybe, I still have some reservations.  I think that in
an optimized implementation this could add
significantly to the CPU overhead.  Consider the JPEG
file example using an API similar to the one I
suggested to John Barton.  In a careful implementation,
there is near to 0 CPU time per byte, if the
implementation DMA's the file in from disk and DMA's it
out onto the network.  That's the classic optimization
done in fileservers and it seems like a good thing to
be able to support.  If you have to do even a simple
checksum, then you've got CPU time per byte.  You also
quite concievably have lots of cache polution in the
processor caches, as you are streaming megabytes or
gigabytes of data through the CPU.  I'm not saying this
is necessarily impractical, I'm saying that it's far
from obvious that the impacts are negligble, especially
given that part of the reason for MTOM is to deal
efficiently with very large amounts of data.

Also: isn't the point of using the IDs exactly so that
you don't have to check the contents?  Whatever the
other merits of using the same MIME part for two or
more bits of identical content, if you're going to
check the content anyway, wouldn't it make some sense
to skip the IDs? 

> I agree that UUID/GUIDs may not be usable in every
> environment but note that RFC 2111 requires that "the
> Content-ID of a MIME body part is required to be
> globally unique" so the problem exists independent of
> MTOM usage. Simply reusing the content-id of the part
> as the attribute value would suffice for MTOM
> multi-reference support.

If I understand what you're saying, that would put
the CID URI in both the hint attribute and also
in the xbinc:include?  Seems sort of strange.
Maybe I'm misunderstanding.

> Regards,
> Marc.

Thank you for your patience with my concerns!


Noah Mendelsohn 
IBM Corporation
One Rogers Street
Cambridge, MA 02142
Received on Monday, 17 November 2003 21:08:10 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 6 January 2015 22:01:24 UTC