Implementation of SYNC_NV bit

Elliott, Robert (Server Storage) Elliott at hp.com
Fri Jul 23 13:54:49 PDT 2010


Formatted message: <a href="http://www.t10.org/cgi-bin/ac.pl?t=r&f=r1007233_f.htm">HTML-formatted message</a>

The same question should apply to the FUA vs. FUA_NV bits in the READ and
WRITE commands.
---
Rob Elliott    HP ISS Platform Technology - Server Storage
From: owner-t10 at t10.org [mailto:owner-t10 at t10.org] On Behalf Of Gerry Houlder
Sent: Monday, 19 July, 2010 4:21 PM
To: T10 Reflector
Subject: Implementation of SYNC_NV bit
My company has been reviewing the definition of the SYNC_NV bit in the
SYNCHRONIZE CACHE command.
The current definition has the plain SYNCHRONIZE CACHE command forces all
volatile and non-volatile caches to be flushed to medium while the setting of
SYNC_NV bit only forces volatile caches to be flushed to medium. Thus, if an
implementation has several Gigabytes of non-volatile cache the plain
SYNCHRONIZE CACHE command could take minutes to complete but the (more
recently defined) SYNCHRONIZE CACHE with SYNC_NV=1 could complete much faster
if only a few Megabytes of volatile cache needed to be flushed to the medium.
Unfortunately, this seems backwards from the host expectations; hosts expect
the plain SYNCHRONIZE CACHE to be completed rather quickly and might be more
willing to accept a long (could be minutes) completion time in a newer
version of the command (like one that has the SYNC_NV bit set).
My purpose of sending this email is to gather opinions from host vendors to
see if:
(a) do you use the SYNCHRONIZE CACHE command with SYNC_NV bit set to 1;
(b) what expectation is ther for completion time of this command with and
without SYNC_NV bit set; and
(c) how would you feel about redefining the SYNC_NV bit to invert its
meaning?
If the SYNC_NV bit were changed in meaning, I would propose to make the plain
SYNCHRONIZE CACHE command flush only volatile cache to medium and SYNC_NV=1
would cause non-volatile caches as well as volatile caches be flushed to
medium.
Unless the responses to this email are overwelmingly negative, I plan to
introduce a proposal at the next T10 meeting that will change the meaning of
the SYNC_NV bit is this manner. I want to hear opinions, both for and
against, ahead of time so I have a sense of how this group will treat this
proposal.



More information about the T10 mailing list