We’re looking at speccing an EC2 instance to record a number of 1080p H264 streams:
4 x 10 Mbps streams
1 x 20Mbps stream
Plus to do trans-rate encoding of the 20Mb stream and act as an http live origin for clients connecting through HLS via Cloudfront.
We’re wondering what size EC2 instance to go for? We’re currently running an m3.medium but we’re getting weird connection drops going in to EC2 when we try and publish a 20Mbps stream.
Should we be running a bigger instance? Or any ideas on troubleshooting / performance testing?
We do not have any performance data on the capacity of the EC2 instances. You will need to perform your own tests and establish the capacity limits for the different types of EC2 instances.
If you will to transcode/transrate the incoming streams, you will need to use the Transcoder feature, which is using up a lot of CPU resources. Because of this, you should look into instance types which have a lot of CPU resources available.
The m3.medium may be a bit lightweight for transcoding plus your hi-def streams. If you check in Wowza Streaming Engine under the Server Tab / Server Monitoring, you should see some graphing showing your current usage of CPU, Heap, Memory, and Disk. Along with network statistics further down the page. You may want to check the utilization on this page and increase the size of your system if the usage is high. (Note: Total Memory on a Linux System will almost always be high, since Linux uses available memory for caching but your Wowza Heap usage shouldn’t be close to 100%)
Wowza does offer a load testing tool but it’s typically used to determine the max client load. You can obtain the Load Test Tool here.
If you look at your [install-dir]/logs/wowzastreamingengine_error.log & access.log files, do you see any error messages related to your dropped connections? The drops may be unrelated to your EC2 instance size.