Post

1 follower Follow
0
Avatar

Compute overhead on AWS

Is an additional overhead for compute on AWS, or do the standard server calculations hold.

For a c4.2 large conferencing node my calculations are that it should provide for 19 HD calls.
(2.66 x 8) / 1.1 = 19

Is that approximately right?

Simon Downey

Graham Walsh

Please sign in to leave a comment.

1 comment

0
Avatar

Hi Simon,

The 1.1 normally comes with a 20% overhead so normally you would do GHz over 1.3 to 1.4 for most E5-2600 CPUs, assuming BIOS settings are set to full performance.

Experience in AWS shows that the c4.2xl provides 13-15 HD. BIOS would most likely be set to Balanced on these hosts in AWS, so a bit lower result then you would expect if you happened to have the same CPU onprem (that exact CPU is for the time being not available on the open market so the above is a slight assumption).

However even if it has a bit lower capacity, Pexip nodes in AWS have the beauty of being able to be powered on and off. Today you can do that manually or trough Pexip and AWS APIs, and in our upcoming release we are planning to automate this. Meaning it can power on additional AWS nodes (that you pre-deploy then shit down) when you are close to run out of capacity, and then power then down when you do not need them any more.

This benefit in my view outweigh the pure performance as you can spin up many additional nodes as needed – automagically.

Reach out to your local Pexip SE or sales rep to see a demo of this today.

Marius Nilsen
Pexip

 
Graham Walsh 0 votes