[mythtv-users] my experience with myth on a solid state drive
rongblor at googlemail.com
Thu Jan 5 18:26:05 UTC 2012
On 05/01/2012 16:20, Ronald Frazier wrote:
> A few months ago, I got a new larger SSD drive for my desktop
> computer, to replace the older, smaller SSD I had previously (a 1st
> generation Intel 80GB). I then decided to take that old drive and use
> that as the primary drive for my myth system. Wow, what an incredible
> performance boost it has been.
I did a similar thing about 9 months ago and bought an SSD for my myth
backend / server and I agree. It's one of the best performance
improvements you can make at the moment.
> Before I made the switch, I did a lot of reading about SSD drives in
> linux, and people talking about putting /tmp or /var or this or that
> onto a physical drive or a tmpfs. The reasoning for all of this is
> that SSDs wear out from excessive writing (unlike HDDs). I figured I'd
> ignore all of that advice and just give it a shot with everything all
> on the SSD (except for my recordings/movies, of course). The only
> thing particular I did was to pay attention to partition
> alignment...some resources said to align it to 512K boundaries, other
> said 1MB boundaries. I played it safe and aligned it to 1MB.
I too have taken the approach of just dump it all on the SDD and not
worry about wear leveling, etc too much on the basis that by the time my
SSD wears out they will be dirt cheap and I can just replace it.
> The intel drives are rated for at least 5 years of 20GB/day of writes.
> You can check the current wear level using the SMART feature of the
> drive. After using this drive in my Windows box for over 3 years, the
> drive still had 93% of it's rated life remaining, so I should have at
> least 4 years of life at 20GB/day. So I began running my system and
> periodically monitoring the wear level. Immediately I discovered I was
> writing about 30+GB/day, which meant the drive would only last about 3
> years. So off I went trying to figure out what was doing so much
> writing. Was it mysql, or logging, or the fact that I'm also running
> my 2 frontend systems (which are diskless boots) from this drive, or
> /tmp? So I began systematically moving things one by one to another
> partition, watching it for a few days, and then moving back when I saw
> no change. Finally when I got to moving /tmp, that made all the
> difference. SSD writes dropped down to about 2GB/day, which means it
> would take 40 years to use up the drive's rated write life.
> So what was doing all the writes to /tmp? Several times an hour, myth
> recalculates its schedule. When it does, it uses about 500MB of
> temporary mysql tables. That alone was responsible for over 90% of the
> writes on the SSD. So instead, I created /tmp as a tmpfs partition.
> I've got 2GB of ram, so that gives /tmp up to 1GB to use.
> So far that's worked just fine with just that one change. I never had
> to mess with all of the other things people say to do to minimize
> wear, like alter the io scheduler or the cache flush timings, or move
> /var or /var/log.
Thats interesting. How exactly are you montoring wear level and the
amount of data being written to your SSD? Is it specific to the Intel
ones? - mine is a Kingston I think
Sounds like I might need to think about switching mysql temp tables to
ram rather than SSD as well if it's a simple change.
More information about the mythtv-users