[mythtv-users] Combined FE/BE using USB for all I/O?

[Simon Hobson]

Eric Sharkey <eric at lisaneric.org> wrote:

>> I wouldn't.  At some point (i.e. some number of simultaneous recordings)
>> you are going to run into spindle-contention where having more separate
>> drives will help.
> 
> http://en.wikipedia.org/wiki/DVB-T
> 
> "Allowed bitrates for the transported data depend on a number of
> coding and modulation parameters: it can range from about 5 to about
> 32 Mbit/s"
> 
> That's 4MB/s for the high end, or 12 MB/s for three tuners recording
> max-bit rate.

It's not the bit rate that kills them.
As Brian points out, there's all the seeking as well. If you listen to a drive while one recording is going on, you'll hear it "tick" about once a second. That's it doing a "seek - write video stream - seek - write metadata - ..." cycle. At some points, there may be more than one seek to write the metadata, and at some points there may be more seeks to read/write database tables if you haven't put them on another disk. As you increase the number of streams, so the number of seeks goes up.
Then in amongst this, add the "seek - read - ..." cycles for each process that's reading data (whether a frontend playing a recording or a job comm flagging or transcoding) and there does come a limit - typically waaaaaay below the throughput capability of the drive if you just look at transfer rates.

Perversely, having a high end processor can make the situation worse - if it can transcode or commflag faster, it'll place more demands on the disk than would a slower processor.

My first backend was "low spec" (too low really, but it was all I had at the time). It could 'just' manage two recordings and one playback - most of the time and if I ran no user jobs. Even so, many of my recordings has a short period of "hiccups" a certain length of time into them - as the burst of database activity when another recording started/ended tipped the balance to the "can't quite cope" side for a few seconds.

Having 2 disks doesn't eliminate the problem, but statistically it dramatically reduces the likelihood of hitting the limits on either disk.