Suggestion for instance selection

[rahul7827]

Hello Team

We are going to start clustering on AWS. I am thinking to proceed with g4dn.xlarge as origin in auto scaling and c5.xlarge as edge server.

While checking the pricing i found this.

 

At a slight change in pricing, benefit is memory is doubled.

However both instances having their own advantages, one is "cpu optimized" and other is "memory optimized".

What do you recommend which one should i proceed with.

[Burak]

Hi @rahul7827,
For WebRTC I advice CPU forever. But also I can give you a tip.

You may use Graviton2 (ARM based) instances. Their performance are similar but they are cheaper.

These are Intel processors.

 

Last but not least, I don't know why but Gravitons are even cheaper in Mumbai than other regions.

 

[rahul7827]

Thank you for your suggestion @Burak

As earlier you recommended Graviton2 and I followed that and in our staging environment one of my colleague is setting up this instance in Mumbai region. This is really cheaper.

As per our client location, we have to keep the production environment in their region also. So I had a confusion which one should i go with. This is the first time i will use graviton so having less confidence 🙂

I think, I should setup both these instances and put the load and observe the performance.

Also Thank you @Yash for sharing the load test tool and the documentation.

[Yash]

Thanks @Burakfor sharing the information.

@rahul7827
Graviton 2 (ARM-based) instances from the c6g instance family are the same as I showed you in the comparison results yesterday. Let me also share our test results here:

The tests were carried out a year before. 

Load tests (One 480p stream with 100 viewers) were performed on each server separately. Because the CPU usage varied during the tests, we took the range.

After 100 viewers were loaded, the CPU load for c5.xlarge (Intel based instance) was between 30-35%.
After 100 viewers, the CPU load for c6g.xlarge (Graviton ARM based instance) was between 35-40%.
After 100 viewers, the CPU load for c5a.xlarge (AMD based instance) was between 40-45%.

I hope this helps.