Concerns about volatility of open zone resources
gerry.houlder at seagate.com
Fri Jul 31 12:15:16 PDT 2015
Formatted message: <a href="http://www.t10.org/cgi-bin/ac.pl?t=r&f=r1507310_f.htm">HTML-formatted message</a>
Concerns have been expressed by customers about the volatility of âopen
zone resourcesâ, specifically the write pointer. Conventional disk drives
(and other storage devices) do not have this risk of data loss if write
caching is disabled. Many customers (especially in the enterprise and near
line markets) do not run with write caching enabled to avoid this risk of
data loss. This is seen by those customers as a big advantage in terms of
not needing to do journaling or use other techniques to mitigate the risk
of data loss from unexpected power loss.
The specific cases where open zone resource persistence is needed are:
a) successful write commands when write caching is disabled;
b) successful write commands with FUA bit set to one;
c) successful write commands (even with WCE=1 and FUA=0) for which a
subsequent SYNCHRONIZE CACHE is successful.
Some of these customers are now very concerned that zoned block devices
have no way to avoid the risk of data loss because open zone resources are
always described as volatile unless the zone is changed to closed or full.
Now these customers are faced with a difficult choice in order to use zoned
block devices in their systems:
a) redesign their systems to add journaling or other data loss
mitigation techniques; or
b) send CLOSE commands or SYNCHRONIZE CACHE commands frequently to limit
the risk of data loss.
Neither of these options is very appealing in that they add a lot of cost
(e.g., more storage devices to do journaling, etc. that are not subject to
data loss due to unexpected power loss) and/or lose performance (due to the
processing of extra CLOSE commands and/or commands for journaling). They
want to have an operating mode (like conventional drives with WCE=0) where
write data is persistent if the write command ends with GOOD status.
I am preparing a proposal to require open zone resources to be persistent
if write caching is disabled, write commands have FUA=1, or cache
synchronize operations are successful. i hope to have it available for
discussion at the Aug. 3 telecon, if time is available.
More information about the T10