RE: A question on Zero Copy

[Douglas Otis <dotis@sanlight.net>]

David,

In the prior note, I did agree with Y.P. with respect to a need for a
content directed placement of data if there are no other provisions within
dedicated appliances.  As such, I did agree with you and was only attempting
to defend Julian's positions that such content directed placement would not
be needed in a dedicated target.  RAID hardware is often designed to use the
Fibre-Channel frame to allow this type of content directed placement.  For
specialized hardware that does post reconstruction of FC frames from an
encapsulation would involve but a few pointers and associated lengths easily
contained within a small ASIC to allow simple post conversion.  iSCSI
methods are less able to use such simple post processing techniques as the
lengths and number of pointers would be less bound.  The length restrictions
of FC frames also makes for an fixed length technique for frame discovery
also not easily done with iSCSI.  One could simply place an FC frame at
fixed intervals.  Buffer to buffer management would also be made far
simpler.  If the argument is about what is already available, then
Fibre-Channel encapsulation becomes the obvious choice and iSCSI adds only
complexity.

Doug


> With my co-chair hat off:
>
> > If you imagine even RAID controller devices have memory management that
> > operate in a manner to resolve data discontinuities and header structure
> > placement, then post mapping replaces look-ahead filtering.  Such post
> > mapping has greater freedom.
>
> One can imagine what one likes, but that doesn't mean that all
> devices will
> be built that way.  That sort of memory management has both hardware and
> software implications that implementers may choose to trade off for other
> things.  YP's description (allocate buffer, issue R2T to fill it) is a
> perfectly
> reasonable way to go about building a RAID device, as there are products
> on the market today that operate in precisely that fashion.  As
> far as I can
> tell, the issue of whether initiators and targets need to be able to put
> inbound
> data into pre-determined locations in memory is implementation dependent
> and depends on a variety of factors that involve software and hardware
> architecture/design issues.
>
> --David
>
> ---------------------------------------------------
> David L. Black, Senior Technologist
> EMC Corporation, 42 South St., Hopkinton, MA  01748
> +1 (508) 435-1000 x75140     FAX: +1 (508) 497-8500
> black_david@emc.com       Mobile: +1 (978) 394-7754
> ---------------------------------------------------
>
>