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

[Jean-Yves Avenard]

On 19 August 2014 05:56, Eric Sharkey <eric at lisaneric.org> wrote:
> Yes, it mostly does, thanks.  What you describe sounds pretty good to
> me.  I still don't like ring buffers in general, but as long as the
> writes are asynchronous and the sync's don't rely on the ring buffer
> it really should be ok in practice.

ringbuffer because it's the easiest and most portable approach.
remember that myth runs on various platforms (BSD, Linux, OS X,
Windows), not all posix compliant, that the core code is almost a
decade old and only required updating recently for minor fixes.

>
> But can you explain why multiple people on this list seem to think
> that I/O contention is a serious problem, such as Stephen
> Worthington's "All it takes is one moment when the
> disk is too busy and you will lose a block of data and seconds of one
> or more recording."?  That just doesn't seem to be the case.

because that just can't happen.
Assuming the disk is slow or the system can't write to disk quick
enough, once the write buffer is full, new data can't be added, the
whole recorder will stop writing new data to disk.
the recording will be truncated and no more data will be written to
it. It can't be a case of partial data missing, it would be *lots* of
data missing.

So I'm suspecting that the issue is different, the analysis of it
being just wrong, and that it's not due to the threaded writer.

The other way I think could happen, still within the recorder but not
the writer. As I mentioned earlier, data coming from the V4L adapter
is continuous and isn't buffered (well except the very little the
driver will buffer). If there's no listener the data is lost. It's a
bit like how serial ports work

So if the system is unusually slow, if the V4L adapter is virtual
etc... that by the time the system is free again to read from the V4L,
some data has been lost, or that data is just lost regardless on the
V4L interface.

I can't think of any other way that would explain random, partial loss
of data: nothing to do with how myth writes to disk with those
symptoms.

>
> So far, the only sensible reason to prefer 2x1TB drives over 1x2TB
> drives is "I have a couple of 1TB drives I can recycle".

2x1TB does give you higher IOPS, as you can write to two disks at once.
I/O throughput is rarely a point of contention with DVB/ATSC recordings.

plus if one disk dies, you only lose 50% of your data (assuming you
properly use storage group and not some RAID0 setup)