Hello,

On 20 Aug 2013, at 17:39, Romain Beauxis <[email protected]> wrote:

> Hi all,2013/8/20 Ashworth Payne <[email protected]>
> > Stream to one server and have the other server relay it?
> 
> This is indeed the best solution I can think of right now. icecast relays are 
> designed to be backup streams, I believe, so they seem like the appropriate 
> approach to me.
> 
> Romain


We are in a particular case.

Today we have several thousands of LS pushing streams in a single master. This 
master is connected to 8 slave servers on which listeners are connected to. We 
are using the icecast master-slave system to do that.

When a station doesn't have listener we stop it to avoid CPU time waste. We are 
starting and stopping thousands of stations thousands times a day. So we add 
and remove mountpoints frequently. We have developed a process that could allow 
a slave to find a station not present on the main master to grab it from a 
secondary "master" server. If the main master doesn't run the station it is 
because it is in troubles (ie it is crashing) so relaying this server to the 
secondary master is useless as the server is in trouble an unable to serve 
slaves (so it's unable to server a relay server).

So summarise :
- 10,000 liquidsoap instances are pushing stream in an icecast server (called 
the main master)
- 8 front-end icecast are connected on this main master delivering the 10,000 
of stations to up to 160,000 listeners

What happens if the main master is in trouble (memory leaks, stats link down, 
motherboard failure, ….) ? The 160,000 listeners are deconnected from the 
slaves as the main master is a SPOF.

The idea is to allow the slaves to connect to another master having the same 
streams available (the automatic switch over is done and working), but in this 
case, we need the stream to be present on that server (that could not pick it 
up from the dying master).

Using real mountpoint+fallback on the slaves instead of the master-slave system 
is not usable as we should set up to 120,000 mountpoints on it (all the 
stations - playing and not playing one) and icecast doesn't like that. We are 
adding more than 100 stations every single day. So we should add them and -HUP 
icecast a lot of times during the day.

So you would say, use more master and split your stations on it. But as a 
single slave could only have a single master my bandwidth sharing between 
servers will not be as efficient as today. As I know every slave as the whole 
mountpoint list, I could loadbalance very well the listeners on my slaves.

The easiest way for us is pushing twice the stream in 2 different masters 
directly from LS of via a local service on the liquidsoap server (it will be a 
SPOF at liquidsoap server level, but it will not impact whole the stations at 
the same time).

Today, a single server (0,5U) is able to use 260 liquidsoap instances.

TIA,

Jef
> 

Attachment: signature.asc
Description: Message signed with OpenPGP using GPGMail

------------------------------------------------------------------------------
Introducing Performance Central, a new site from SourceForge and 
AppDynamics. Performance Central is your source for news, insights, 
analysis and resources for efficient Application Performance Management. 
Visit us today!
http://pubads.g.doubleclick.net/gampad/clk?id=48897511&iu=/4140/ostg.clktrk
_______________________________________________
Savonet-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/savonet-users

Reply via email to