[BBLISA] SSD's - really any better performance?

[Edward Ned Harvey]

> We have been experimenting with ssd's for mysql servers.  Though I
> don't
> have hard numbers yet, the performance difference has been night and
> day
> (vs SAS).

I'm curious about this - because the numbers I found wouldn't suggest a
night-and-day performance difference, except during random small reads.

Do you have any knowledge of your usage pattern?  Are you doing mostly
random small reads?  Perhaps my numbers are somehow not well representative
of actual performance?

Have you run any iozone, or similar benchmarks on it?


> So far, I am very  happy with the performance but I am still
> concerned on MTBF.  Like you say, it's different than 'regular' drives
> and I don't know anyone with personal, long term experience with ssd's.

I also don't have personal long term experience, but I work for a company
that produces some chips for SSD drives, so the engineers around me all know
a lot about it.  Here's what they say:

By using techniques such as write remapping (which has different names based
on manufacturer) and by overallocating the storage medium, they're able to
keep the write operations pretty well spread out across the medium, and also
able to ... essentially predict and correct bad blocks before they happen.
They say that because of consumer fears such as yours and mine, they're
overcompensating, and making the SSD's generally more reliable than regular
hard disks.


sdfdsfdsf