Recent Topics Added
Hello red5 inventors and red5 users,
we are running red5 latest release with high-perf setting in production. We expect 20K concurrent users watching various streams from single production box (hosted on vps).
Facing some performance issues
1. red5 stop crashing after few hours
2. video keep buffering etc.
Would like to get more info on following:
1. how many concurrent users red5 can support assuming decent production box (vps hosting with 3GB RAM)
2. Any test cases/scripts that would help load test red5. some pointers/strategy on how to proceed with load tesing red5
3. Am sure many using red5 in production. Want to get an idea of what load balancing strategy they you guys follwoing like what load kind of load balacer you using, some statisticks like how much users your environment supports, any pointers to tweak red5 performance.
Page 1 of 1
Red5 load testing/performance tuning
performance issues
performance issues
Quote
20K concurrent users watching various streams from single production box
Wow! What data host do you have that provides you 2.384 Gigabits per second bandwidth and what kind of network IO hardware do you have? That's incredible.
1) Depends on bandwidth of each stream. Lets say they are 350kbs - 700kbs VBR. Last load test I was part of, we settled on 350 users per VM slice on 'middle of the road' hardware. This number gave us the M.T.B.F we felt that was large enough that we could turn off the cell phone over the weekend. Many admins may settle on higher numbers. At 450 user it would fly for awhile. Smaller KBS streams may provide higher numbers also. This was a couple or Red5 versions ago. If you were to just use shared objects, you could easily provide support for several thousand concurrent users.
2) Adobe produced a C binary that can run on linux/windows to load test a server. the Adobe load Test tool. Or you can roll your own.
3) Load balancing in case '1' was done by a php library that was provided by the data center. From the id of the machine the library gave us, we knew which VM was the least loaded. Other admins create custom gateways that communicate with red5 directly, since they figure some usages are heavier than others, such as publishing versus shared object handlers.
Using our desired MTBF numbers, you will need around 60 VM slices. Less, depending on the content bandwidth and how much seeking will be done.
NOTE: Our numbers are based on a single broadcast stream, Live, not files. We use the built in StreamProxy class in red5 to forward the stream to our Slaves.
-
dominick
Administrators
- Group: Administrators
- Posts: 176
- Joined: 01-April 11
- LocationBoston, MA
Posted 28 July 2011 - 07:38 AM
#3
lol, what Andy said!
--
Dominick Accattato
Red5 Co-Project Manager & Engineer
---
http://www.infrared5.com/
http://code.google.com/p/red5/
http://code.google.com/p/jedai/
twitter: @newviewnetworks
blog: www.newviewnetworks.com
Dominick Accattato
Red5 Co-Project Manager & Engineer
---
http://www.infrared5.com/
http://code.google.com/p/red5/
http://code.google.com/p/jedai/
twitter: @newviewnetworks
blog: www.newviewnetworks.com
Share this topic:
Page 1 of 1
