[multicat-devel] dvblast over wireguard v.s. multicat over wireguard

[graham lenton]

I am conducting some comparative testing of unicast video streaming on a
private network using a wireguard vpn. The objective is to demonstrate the
low latency and zero packet loss of the VPN under ideal conditions.

To eliminate vagaries of public networks, routers and so, the wireguard vpn
is between two Raspberry Pis running 32-bit Raspbian on an isolated gigabit
network switch with no other workloads or network traffic present. The
network, memory, disk, and cpu resources available are more than sufficient
for the task at hand, and do not approach exhaustion at any point in the
tests described here.

The streams are obtained from UK Freeview via a USB DVB-T adapter, and
streamed using dvblast to two VLC players - one via the LAN, and one via
the VPN, using a multicat instance to turn the stream around to a 3rd
machine running VLC players side by side for comparison.

With live DVB-T single-programme transport stream content via dvblast ->
VPN -> multicat -> LAN -> vlc, the results are excellent, and only a tiny
delay is present compared to the direct dvblast -> LAN -> vlc.

These results are achieved using plain UDP throughout and minimal buffering
in VLC.

With recorded content, which was captured via dvblast -> multicat -> .ts /
.aux file, there is some problem across the VPN which is not present when
playing the same file direct across the LAN.

To be clear - the exact same .ts / .aux file is used for the LAN and VPN
scenarios.

For LAN: multicat with file input -> LAN -> vlc works perfectly.
For VPN: multicat with file input -> VPN -> multicat ->LAN -> vlc exhibits
a considerable degree of breakup, as if there was some packet loss or
jitter.

The problem is completely reproducible and even the break-up artefacts
become familiar after extended testing..

I have tried various combinations of RTP, UDP, -C to rewrite continuity
counters, -P to restamp PCRs, DTSs, and PTSs, -w to send with RAW.

I believe the difference may be in the way dvblast buffers and smooths the
stream. I am using BBC News which is typically somewhere around 2Mbps.

Can anyone offer insight into what might be causing the differences here ?

Thanks,
Graham
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/multicat-devel/attachments/20210927/3225b556/attachment.html>