Edge Expander Set and STP/SATA

Seto, Pak-lung pak-lung.seto at intel.com
Wed Oct 9 07:41:46 PDT 2002


* From the T10 Reflector (t10 at t10.org), posted by:
* "Seto, Pak-lung" <pak-lung.seto at intel.com>
*
In the Edge Expander Set proposal (02-359r2):

The definition of Edge Expander Set is "A set of edge expander devices that
are bounded by a subtractive routing port from one member of the set."

"Edge expander devices may be grouped into edge expander device sets.  The
edge expander device sets are bounded by the direct routing phys attached to
end devices and the phys supporting subtractive routing that are:
	a) Attached to the phys of a fanout device; or,
	b) Attached to the phys supporting subtractive routing on another
edge expander device set; or;
	c) Attached to an end device."

>From the definition of Edge Expander Set, there are no restriction on the
number of level of edge expanders within a "set of edge expander" can
cascade.  It can be configured like a tree structure with many level as long
as it meets the above definition and with no more than 64 end devices.  Or
it can be constructed with a string of 3 port edge expanders of up to 64
levels (practical/meaningful limit) on each side of the Fanout expander, etc
(.  For SAS, it may be OK (with long latency) but with STP/SATA, it can
cause a lot of problem because in STP, the expander provides SATA emulation
within the expander link.  It requires certain amount of buffering
(depending on the expander latency) to satisfy the HOLD/HOLDA SATA flow
control protocol.  For each additional level of cascading, it will require
additional X amount of buffering.  Architecturally, it is a problem, for
real implementation there may be a practical limit of a few levels.  In the
proposal, all the diagrams only show 2 levels of cascading.

In this Edge Expander Set proposal, since there is no restriction on the
number of expander levels in expander set configuration, it will require a
lot of buffering within the expander to anticipate many levels of expander
cascading, it is very costly.

Either the SAS specification sets the limit the number of level of cascading
or individual implementation will set their own limit.  It will be better
for interoperability reason that the SAS specification set a reasonable
limit number of levels.

Pak Seto
Intel Corporation
Intel Communications Group
Storage Components Division
pak-lung.seto at intel.com
978-553-2159

*
* For T10 Reflector information, send a message with
* 'info t10' (no quotes) in the message body to majordomo at t10.org




More information about the T10 mailing list