I recently switched from a small VPS provider to AWS EC2. My traffic usually comes in bursts throughout the day and I therefore concluded that running multiple t2.micro instances would be by far the most economical solution.

For those who don't know about it: micro instances are very cheap and you get 10% baseline usage of 1 CPU core. However, you are allowed to use up to 100% for short bursts if your average use stays below those 10%.

It is therefore much cheaper to run 10 micro instances (0.15$/hr) with potentially up to 10 cores running full load simultaneously for short bursts, than having some similar performance with one large instance (e.g. m4.2xlarge - 0.57$/hr)

The only disadvantage of the micro instances is that they would have to run 10x the base usage for the OS compared to a single large instance.

I currently use Windows Server which already uses a good chunk of the 10% allowed CPU usage. How does ubuntu compare to that? Is it much lighter (esp. without GUI) and would leave more of the performance for the actual hosting tasks?

closed as off-topic by HopelessN00b, Greg Askew, HBruijn Jan 31 '16 at 13:54

This question appears to be off-topic. The users who voted to close gave this specific reason:

If this question can be reworded to fit the rules in the help center, please edit the question.

  • Allow me to paraphrase your question into one about cares. The engine on my Toyota is using too many RPMs. How does Ford compare to that? Is it much lighter, especially without an automatic transmission and would leave more performance for the actual driving tasks? (Hopefully, this illustrates to you the massive, fundamental problems with your question, why it's unanswerable as written, and likely beyond saving.) – HopelessN00b Jan 21 '16 at 15:13
  • @HopelessN00b I thinks it is a pretty straight forward question: How much CPU/RAM is used by ubuntu (with/without GUI) and how much is used by windows server. – Chris Jan 21 '16 at 15:45
  • How much RPM/fuel is used by Ford (with/without automatic transmission) and how much is used by Toyota? Part of the problem is exactly that you know too little about these OSes to realize how awful your question is. For example, your assumption that Windows has to have a GUI. You may as well be asking if the best way to get from London to Beijing is by driving a Ford pickup truck or driving a Toyota minivan. The only correct answer to your question is: "Neither. You're doing it totally wrong, and you haven't provided enough information to tell you how to do it right." – HopelessN00b Jan 21 '16 at 15:57

I have an Amazon Linux server (no GUI), which is somewhat similar to Ubuntu, but not the same. The CPU utilization when the server isn't serving requests is around 0.03%, and I get to maximum CPU credit fairly quickly.

Eight t2.micro instances comes to $86/month, one m4.2xl is $350 a month. If you go with the t2 instances you also need a load balancer, which if you do 50GB/month is around $20 extra (I think). You also need to architect your website so you can serve from multiple web servers rather than one big one, which could mean you need to pay for a RDS database rather than running it on your instance - or run it on another t2. t2.micro has 1GB memory, you might find a large t2 instance works better if your application is memory intensive.

You should probably look into optimization, because needing 8 cores is pretty heavy duty. Caching pages by anonymous users, even for a short time like 5 minutes, has huge benefits. Using a CDN can offload work from the web server. Setting up headers for client caching can help.

If you say exactly what you're serving you may get better advice.

  • Thanks for your feedback. I came to a similar conclusion and launched some instances today: t2.micro/Windows (no load): CPU 3-7%, RAM ~800MB, install size ~20GB Ubuntu (no load): CPU: < 0.1% RAM: 80MB DISK: <2GB ubuntu & LXDE GUI: CPU: <1% RAM: 200MB DISK: 2GB This clearly shows that Windows will not be a good option for the micro instances since it does not leave much of the allowed threshold for the application. I will give it a try with some (3-4) micro instances with a load balander and cloudfront cdn to see how it works. (I use serve a B2B meteor application) – Chris Jan 21 '16 at 21:25

Not the answer you're looking for? Browse other questions tagged or ask your own question.