How many streams with small istance

[newtargetweb]

Hi,

may I ask you a couple of informations about EC2 usage of Wowza?

In particular, if I'm live-streaming a m4v with a 900kbps encoding how many streams I'm able to play at the same time within the instance limits?

[rrlanham]

Conservatively, on a small instance, I would say 150 concurrent for 900 kbs stream.

Expect to get a total of about 150-200Mbps from a small instance and about 300-350Mbps from an xlarge instance (with a large instance) somewhere inbetween.

Charlie

small instance
150000/900 = 166

x-large
300000/900 = 332

Richard

[newtargetweb]

thanks for reply rrlanham

[rrlanham]

Also, one other thing: 900 kb/s is very high for an internet audience. It will work for some users, but it is likely to be a bad experience for a significant number.

400 to 500 kb/s would be much better. For reference, the Wowza sample video "Extremists.m4v" is under 450 kb/s

Richard

[kebabber]

xlarge instance gives you 300 - 350Mbps .

What if you need approximately 800 - 900Mbps for a big live event?

newbie adding 3 xlarge instances would help i guess

[rrlanham]

You can use "origin/edge" configuration. This is where you run multiple edge servers with load balancing. The origin handles the live stream. Clients connect to one of the edge servers. The edge server gets the live stream from the origin.

See the "Multiple Server Live Streaming (Live Stream Repeater)" section of the User Guide.

This is the server-side load balance method:
http://www.wowzamedia.com/forums/showthread.php?t=4637

Richard

[christr]

Richard, do you know if it's possible to cascade load balancers?

Here's my issue. My streaming servers have no problems dishing out 700 - 900mbps over out gig uplinks. But, I've only got those 2 and right now we only have 2gig in our datacenter anyway. (waiting for Level3 to add another but that's gonna be at least 60 days). What I'd like to do is use load balancing and Amazon, but Amazon instances can't be evenly balanced with what I've got.

So my idea was a load balancer that had my servers in the mix, with another load balancer. The 2nd one then sends calls off to Amazon. But I'm not sure if the players can deal with the multiple redirects. Have you or anyone else done this before?


My other option is to try to find another Wowza CDN that's got spare capacity and I can add them to my balancer temporarily for large events. Sent off a few emails to ones I know about tho and never heard back.

thanks,
--Chris

[rrlanham]

Chris,

Why do you say they can't be evenly balanced? They should be evenly balanced in terms of number of connections.

You could make one of your servers be the load balance server, the other be an edge server, and you can add Wowza ec2 servers as edge servers.

You could make all the edge servers only ec2 servers so they would be identical.

Richard

[christr]

I thought all the servers had to be similar in connection capacity? Or is there a way to define how many connections each server could take? If I could give each server their own max connection count that would be perfect. I could say each of mine can take 5000 or whatever and then the amazon stuff can take 1500 or whatever they can handle.

The amazon stuff seems to be able to only run 100 -300mbps each, but my boxes can handle 2 - 3x that amount of traffic. Plus I'd much rather push from my DC then Amazon (much cheaper ).


thanks,
--Chris

[rrlanham]

Chris,

You're right that the load balance system assumes all the edge's are equal, and redirects to the "least loaded server" by connection count, not capacity.

But the latest version of the Load Balance package has source and documentation. There are two interfaces that you can implement to extend the default load balance system:

ILoadBalancerMonitor

• appendToMessage: use to append to the message sent to the LB server from the edge. You could set a connection limit in an Application property, and see if that is reached and append that data to the message.

ILoadBalancerRedirector

• onMessage will be called each time an edge server sends a status message to the load balancer, so that is where you can catch the current load and appended message from the edge server.
• getRedirect will be called each time a request comes in to find the least loaded server, you can use any algorithm method you like to pick the server to send this redirect to