dillon at apollo.backplane.com
Mon May 24 20:15:26 PDT 2010
:On Mon, May 24, 2010 4:42 pm, Matthew Dillon wrote:
:> For a manufacturer-fresh SSD configuring only 32G out of the 40G
:> leaves 8G which the SSD firmware will use to enhance its wear-leveling
:> algorithms, improving the overall life of the SSD.
:What's the wear rate you've seen on the machines with SSD that you set up?
: (assuming that the SSDs are still in there and you've been tracking
:overall I/O) I recall that the wear rate was comfortingly low.
I've been doing a running test for the last few months on pkgbox64
using the 40G Intel MLC SSD. The work load is the daily bulk build
with the swapcache set to cache data & meta-data, plus normal paging
activity from tmpfs operation.
At the moment smartctl shows ~4TB worth of writes and the wear
indicator is at 97%. It started at 99% so presumably that's 2% wear
This gives us (so far) a run-rate of 2TB per 1%-wear which comes to a
durability of around 200TB. This is extremely good.
Intel specs this particular SSD as having a durability of 35TB (which
is what you get with a normal filesystem doing normal windows-like
activities). The theoretical best-case with MLC flash is a 400TB
durability with a 40G device (10,000 erase cycles).
It takes a month or two for the wear indicator to tick over at current
load levels. Every time it does I get a better feel for the actual
durability. So far I am very impressed with the Intel SSD.
<dillon at backplane.com>
More information about the Users