[OpenIndiana-discuss] SSDs honoring cache flush?

[Richard Elling]

On Apr 24, 2012, at 12:35 PM, Roy Sigurd Karlsbakk wrote:

> Hi all
> 
> There was a discussion some time back about some (or most?) SSDs not honoring cache flushes, that is, something is written to, say, the SLOG, and ZFS sends a flush(), the SSD issues a NOP and falsely acknowledges the flush. 

Some people accused SSD vendors of not honoring cache flush. IIRC, the 
allegations were not proven.

> 
> Now, I've gotten an offer for an SSD that looks good for SLOG, OCZ Deneva 2 C SLC 30GB 2.5" SSD. This does not have supercap, but so long as it honors cache flush, then it won't make a huge difference, since if a write() happens, the SSD cache is flushed to the SSD media, this probably won't take more than a handful of milliseconds, and the difference between that and backing up the cache, seems minimal.

I've known people to use that SSD without issue.

Here is an article that sheds a little bit of light on the various SSD
designs. Unfortunately, it does not address the issue that HDDs have
with their volatile caches and why cache flush commands are so important.

http://www.storagesearch.com/ssd-power-going-down.html

 -- richard

--
ZFS Performance and Training
Richard.Elling at RichardElling.com
+1-760-896-4422