SAS-2: Physical link rate tolerance management QUESTION
Elliott, Robert (Server Storage)
Elliott at hp.com
Fri Jul 6 20:56:29 PDT 2007
Formatted message: <A HREF="r0707063_f.htm">HTML-formatted message</A>
A transmitter honoring "1 per" is fine. That's a subset of the other
rules. Consecutiveness is not implied by these rules.
Expanders cannot guaranteed the "1 per" rules while forwarding dwords in
connections from one physical link rate to another, so receivers are
required to tolerate the fact that they might show up nicely arranged as
"1 per" - they could be bunched together over wider ranges (e.g., four
together, then 508 non-deletable primitives).
---
Rob Elliott, HP Server Storage
elliott at hp.com
________________________________
From: owner-t10 at t10.org [mailto:owner-t10 at t10.org] On Behalf Of
Stephen FINCH
Sent: Thursday, July 05, 2007 2:39 PM
To: T10 Reflector
Subject: SAS-2: Physical link rate tolerance management QUESTION
In section "7.3.2 Phys originating dwords", Table 115 lists the
rate tolerance insertion rate as:
"1,5 Gbps One deletable primitive within every 128
dwords
3 Gbps Two deletable primitives within every 256
dwords
6 Gbps Four deletable primitives within every 512
dwords"
My question:
If an implementation inserts one deletable primitive every 128
dwords when running at 3 Gbps or 6 Gbps, is that phy compliant or
non-compliant with this requirement?
I can find no requirement that the two or four inserted
deletable primitives be consecutive. If they are required to be
consecutive, then I think the standard should say so.
If they are not required to be consecutive, then why not just
state "One deletable primitive within every 128 dwords" for all speeds?
Or are we purposefully allowing transmitting phys to hold off the
insertions and then do a burst? To what advantage? A disadvantage is
the need for (a small) amount of additional space in a speed matching
FIFO if the distance between deletable primitives can be 508 dwords (at
6G).
Steve Finch
STMicroelectronics
More information about the T10
mailing list