Technical blog and Consulting Services
Wowza CPU Performance comparative for Live Streaming sites
The idea of this post is share some technical details about different Wowza servers settings, running into Live Streaming sites.
This 4 years working into multiple projects allow to me recollect stats about performance, currently support more than 100 Wowza Server daily and this list of CPU’s is the common used.
One normal question from my customers are, what server need to buy for my streams?
For this reason into this post I’ll share some metrics and performance graph from multiple servers.
All the server run, with Wowza Media Server 2.2.3 with JDK 1.6.0_XX over Centos 5/6 64 bits.
The list of server for this comparative are: (benchmark from http://www.cpubenchmark.net):
- Intel Xeon X3210 @ 2.13GHz 2,622
- Intel Xeon X3220 @ 2.40GHz 3,046
- Intel Xeon X3430 @ 2.40GHz 3,951
- Intel Xeon X3470 @ 2.93GHz 5,987
- Intel Xeon E5-2620 0 @ 2.00GHz (Not listed into cpubenchmark.net) but is the most powerful server with 6 cores
Methodology of the metric recollections:
- At the moment to take the metrics all the server need to be using 1Gbps during almost 15 minutes (this metrics are based into 30 minutes samples)
- All need to have similar concurrent users connected (the average is 4610 Concurrent Users into this comparative)
We’ll use 5 metric to compare:
- CPU Available (Average)
- JVM Memory Used (Average)
- Server RAM
- Load (Average)
- Connected Users (Average)
Hi Alejandro,
This is great information. For the graph where you show concurrent users, what is the bit rate of the content being streamed? Given the number of users appears pretty consistent, is the upper limit just the bandwidth? In other words, if you were using 10 Gbps NICs, would the number of concurrent users jump significantly on the higher-end servers?
-Chris
Chris, the limit here was always the network card, commodity hardware normally for the prices we paid are limited to 1Gbps.
Only with X3210 we reach the limit of the hardware (CPU) with the other server all have resources available for reach more users.
how do u manage the hdd io bottleneck? 5000+ users would need high random io access. Using SSD or some raid?
My question is similar. how can a regular 500Kbps stream have 4610 concurrent users under 1gbps nic? 4610 here is average concurrent users over a period of time?