>Date: Thu, 13 Oct 2005 08:44:19 +0200 (CEST)
>From: Patrick Boettcher <patrick.boettcher@xxxxxxx>
...
>Hi Matto,
>
>On Wed, 12 Oct 2005, Matto Marjanovic wrote:
>> o The HD5000 can do TS stream demultiplexing in hardware, correct?
>> What kind of impact does this have on its performance (i.e. what
>> kind of difference does this make, and when)?
>
>The HD5000 does not do demultiplexing, but PID-filtering. Ie. in the first
>instance the card just passes the TS packets to the driver, which have the
>requested PIDs. Demuxing is still done in software.
Yah, I think I get it now. Is this summary more or less correct?:
o An ATSC broadcast delivers a fixed-rate MPEG-TS bitstream at ~19 Mbps.
(And a QAM-256 cablecast delivers double this, ~38 Mbps.)
o The TS is a series of packets labelled by PID. It can have multiple
A/V programs, multiplexed together --- each such 'subchannel' has
one or more PID's allocated to it.
o The capacity of an ATSC broadcast is roughly two 720p programs or
six 480p programs. (WGBH in Boston, for example, provides one
720p HD subchannel along with 3 SD subchannels.)
o Unused capacity is filled with null packets (to maintain the fixed
bitrate).
o So, without hardware PID filtering, a tuner card is pumping a
continuous 19 (or 38) Mbps through the PCI bus --- and typically
half of that (9.5 Mbps) is either null packets or some other
program material. These unwanted packets are then dropped in
the bitbucket by the driver.
With hardware filtering, these packets never make it to the bus.
But PCI handles ~1Gbps, right? So I suppose this waste probably isn't
so bad, even with multiple cards --- as long as the packets can be
dropped efficiently on the CPU-side.
-matt m.
