[mythtv-users] Raspberry Pi suitability for MythFrontend

[Raymond Wagner]

On 2/20/2012 03:01, tortise wrote:
> A frontend typically only runs for the viewing time, the backend runs 
> all the time... I think the most power saving is actually to be gained 
> if the pi ran as a master backend

Oh dear god no!  Read the first post I made at the very top of this 
thread again.  The ARM11 in the Pi, based off the ARMv6 architecture, is 
an in-order superscaler processor, with a relatively short pipeline and 
no speculative operation.  While all that simplicity means very low 
power consumption, it also means very low performance.  Branch 
prediction, memory prefetch, and out of order operation are what let you 
crank up the IPCs on a processor, and make that superscaler 
configuration worthwhile.

In other words, that means on average, with code not heavily optimized 
for that architecture, the 700MHz ARM11 will run somewhere around the 
equivalent performance of a bottom end Pentium II.  Now you can avoid 
the jobqueue, and not run commflagging and transcoding, but you still 
have to deal with the scheduler.  Take my backend for example.  I have a 
fairly average system, with two dozen broadcast channels, three digital 
tuners, and 17 active recording rules.  On my 3.3GHz Phenom II, 
scheduling passes take 1.5-2 seconds typically, spiking to 6-7 seconds 
under load.  The scheduler is single threaded, so multiple cores make 
little difference, but the Phenom is still going to be 10-15x faster 
than the ARM.

Under load, you're going to be looking at scheduler times up to a couple 
minutes.  The system will be under load when a frontend is in use, or a 
slave backend is recording.  If a change in data triggers a rerun of the 
scheduler at such a time, it will be disruptive of in-progress 
recordings, and will delay the start of new recordings.  A small system, 
with limited guide data, limited channels, and few tuners can be managed 
by a low powered machine such as that.  The problem is that MythTV is a 
hobby, and hobbies tend to grow over time.  The low performance of the 
Pi and SheevaPlugs is going to put them below the needs of most MythTV 
users.

> I recall there was hope that frontend processing could be done with 
> vdpau and 256M however I've not seen anything about that recently, has 
> that died a natural death?

That discussion was about VIDEO memory.  The VDPAU implementation in 
MythTV was not happy with video cards that only had 256MB of memory.  It 
wanted video cards with 512MB, twice that found in the higher end Pi, 
not to mention however much memory the PC that housed that 512MB video 
card had.