[Wylug-discuss] Got me an fusion-io SSD, want one? Should be screaming fast

[Tom Hall]

>> If anyone else is interested in making stuff go quick I could drop
>> round and do a test with it, see what we get. I should be able to get
>> a nice price on them, they claim 120'000 IOPS which is equivalent to
>> 600+ harddrives and I think the 80G version is about 3 grand.
>
>
> Interesting.  However as part of your testing (after doing the other parts
> as this is likely to permanently degrade it), could you try completely
> filling it, deleting the filler back off again, and then repeating your
> performance tests.  The idea of this is to force all the blocks into use so
> that you then need to erase blocks before they can be allocated.  This
> should put performance into the toilet (comparatively) unless the
> firmware/controller is very good at managing performance.

That is a fairly naive assumption as the performance drop would only
happen with the dumbest of controllers and one write cycle would only
permanently degrade an already broken cell. Most controllers will
pre-erase cells that are not in use and use wear levelling to avoid
killing particular cells. The fusion-io devices in particular have a
user configurable % free that it ensures is always ready to be
written, it "scrubs" (I cant remember the term they use) the whole
bank of cells every 10s (with a noticeable, but predictable and
supposedly negligible, momentary increase in the latency of the
device). As the cells are pre wiped and as I understand it they have a
sort of log rather than metadata it writes the whole device in one
pass then starts over on the pre-wiped part. Random writes can then be
as fast as sequential as long as the % you reserve is high enough.
They seem to have done some good work, for instance I assumed the card
would present itself as a SCSI controller with a disk on but they have
a custom driver the presents it directly as block storage and they
sidestep any legacy protocols (annoyingly for me it does not work with
the version of VMware we are using in work, but does work in with Xen
and is well supported in the Linux (dont know if it's in the mainline
kernel though) and with microsoft drivers)

> After your first period (month maybe) the write performance you will get
> from an SSD will be strongly biased by the erase performance.
>
> Theres a good article on SSDs at
>        http://www.anandtech.com/printarticle.aspx?i=3531
>

It is an OK article, I think when you get to "Putting Theory to
Practice: Understanding the SSD Performance Degradation Problem" you
can simply imagine the wiping of the entire block done in advance in
the background and not have the problems he describes though.

Tom