My company asked me to evaluate a possible server architecture hosted on rackspace to provide live one to many streaming services.
The main idea is to provide a fully "on-demand" service to our customers: we'll use a wowza daily license model, and the same for rackspace infrastructure. So we need to instantiate the servers infrastructure every time a customer needs to setup a live event.
I planned to use the origin/edge architecture to scale and the load balancing to determine which edge server has to serve the current client.
I need to determine the optimal rackspace server configuration for both the origin transcoding/transrating server(s) and the edge servers, assuming that a single inbound stream will be 1280x720@30fps and has to be re-streamed at the same resolution to n clients.
I'm currently evaluating these two configurations:
- RAM: 15 GB - vCPUs: 6 - Disk Space: 620 GB (RAID10) - Public Network:200 Mbps - Internal Network:400 Mbps
- RAM: 8 GB - vCPUs: 4 - Disk Space: 320 GB (RAID10) - Public Network:150 Mbps - Internal Network:300 Mbps
For each of these two machines above I need to determine:
- the number of inbound streams is capable to transcode
- the "n" number of concurrent clients can serve if instantiated and configured as edge server
About the number of transcodings I read this benchmark: Wowza-Transcoder-AddOn-Performance-Benchmark and assuming a 55% CPU max load, I can compare the "Server 4" mentioned in that benchmark with the 15GB Rackspace server above, so it can transcode only 1 stream at a time. Is it correct?
What about the n concurrent clients in case the same 15GB machine is used to re-stream the live video mentioned above? Assuming an average bitrate of 1Mbps for each client and a 20% overhead: public network bw 200Mbps - 20% = 160Mbps corresponds to 160 concurrent users for each edge server. Is it correct?
The idea is that this infrastructure has to scale dinamically instantiating a new edge server when the n number for the current server is going to be reached.
Any advice/suggestion would be really appreciated.