FreeBSD and SSD drives
dave at g8kbv.demon.co.uk
Sat Feb 12 12:36:02 UTC 2011
On 11 Feb 2011 at 13:33, Adam Vande More wrote:
> On Fri, Feb 11, 2011 at 12:57 PM, Chad Perrin <perrin at apotheon.com>
> > Ignoring the TRIM issue for a moment . . .
> > You're probably best off saving SSD storage for cases where you have
> > lots of reads and little to no write activity, unless you enjoy
> > buying new SSDs a lot. Actually, let's not ignore TRIM; the
> > work-around for lack of TRIM support on some drives is a "garbage
> > collection" routine that exacerbates the problem of having to
> > replace your SSDs more often if you do a lot of writes.
> > I guess I would only use SSDs on servers in the same cases where I
> > would let myself be talked into using MySQL -- cases where you just
> > treat it pretty much like a read-only data store, and do not have to
> > (safely) add or change data stored there most of the time.
> Modern SSD's can do a *lot* of writes, wear-leveling and other
> tecniques allow SSD's to be implemented for nearly any workload.
> There's a great deal of literature and facts on this topic if someone
> was motivated enough to research it. Some legends are better off
> fading away.
> Same thing is sort of true with TRIM, on most modern drives lack of OS
> TRIM support isn't the performance hit it used to be although still
> Adam Vande More
Define "a *lot*". If you look up the spec's on the common (currently)
available SSD systems, it's only in the 10's of 1000's writes. Pittiful
compared to magnetic media.
The way they work too, if you write one "sector" you actualy re-write a
much larger block of memory. Wear leveling, not that common with SSD
Hard Drives, but very common with USB (Flash) memory sticks, only goes so
SSD's have a place, but not for things like swapfiles or working data
that changes a lot..
More information about the freebsd-questions