I have never been worried about scaling meteor apps and each time I hear the scaling questions about meteor I feel it one of those buzz questions and in truth not a smart question to ask in most context. My thoughts are clear on this in the followng post.
But lately my biggest concern with Meteor and this might not be a full meteor issue but more about websockets and reactive applications. Is cost at scale. I believe everyone has been ignoring this. But maybe this isn’t a big deal and every language, framework or tech is similar. Why I’m asking about this here.
The average VM depending on your code can handle anywhere between 150 to 350 concurrent users. So if you have 10,000 concurrent users which I believe is not crazy consumer scale yoru looking at the following
I’m using Modules.io as Meteor Galxey in my point of view is not for startups and not realstic I think they missed the ball here.
Using the 512 RAM servo size which is $28.80 per servo and $288 for 10 so if each servo can handle 350 current users (10,000 / 350) I need about 28 servers for simple sake let round up to 30 that would be $288 * 3 = $864 per month.
And let’s look at factor of 5 that would be $4,320 per month
I wonder if this is ok and in pair with other languages / frameworks?
I know from my previous start which was Ruby on Rails with Angularjs front end we were able to scale to millions of requests at the cost of $1,000 to $2,000 per month (to clarify that was on Microsoft Azure, VM Linux. We had replicate set for Mongo primary and two secondaries and then for the app a load balancer and two machines plus running and another for backup if one went down. For the client just one machine and running mode of the static files from CDN).