Vertical scaling - change configuration on the fly.

You cannot vertically scale an EC2 instance while meeting those requirements. To change an instances size, it must be:

1. Stopped
2. Resized
3. Started again

Can you please go into more details on why horizontal scaling is a cost issue? In your example, a t2.2xl is 32x the price of a t2.micro: https://aws.amazon.com/ec2/pricing/on-demand/

There is some loss from the additional OS overhead running on each new instance, but if you horizontally scaled an additional 3 t2.micros (or whatever size instance makes sense for your baseline load), then it would still be cheaper than vertically scaling to a 2xl

You can horizontally scale with an EC2 AutoScaling Group (ASG), here's a few caveats to keep in mind:

• You'll need a way to route requests to the instances. While you could try something like DNS based routing to reduce costs, this doesn't really balance the load, since clients will cache requests. Generally using a load balance, like an ALB is recommended. At lower traffic levels, this would be similar in cost to 2 t2.micro instances
• In production, you'll likely want to have unlimited mode enabled on your T family instances so that bursts don't bring down your application performance. Keep in mind that any borrowed CPU credits will be charged when the instance is terminated by the ASG. If your scaling on CPU usage, then you'll also want Unlimited Mode enabled to prevent issues with scaling based on the CloudWatch metrics (if the credits run out, the CPU metric won't go above the baseline)

As per the other answer: vertical scaling can't be done in the way that you require; there will always be an outage. That's why it's better to design applications so that they can be horizontally scaled and so that they can deal with interruptions between the client and the server.

Taking a slightly different angle here: There are always going to be interruptions to service because (as Werner Vogels says) "Things fail all the time". If you have the flexibility to ensure that the application reconnects seamlessly (from the user's perspective) then you now have the opportunity to scale horizontally which is better all 'round.

Good day. Thank you very much for your recommendations. Analyzing Amazon services, I come to the conclusion that it is very bad that there is no way to change the configuration of the processor and memory, provided that all data remains on the hard drive and works. This is similar to hot swapping a hard drive with a different configuration. This would be an insanely cool thing for any company, not only in terms of financial savings, but also in terms of technology. Thanks again for the replies and recommendations.

Thank you very much for your response. Okay, I'll try to describe the problem more precisely. I have an AWS Nitro Enclaves server and they have a c6in.4xlarge configuration for which my company pays $653.18 each monthly. The work they do is so critical and important that they cannot be replaced by cheaper ones or temporarily suspended. The need for such power is justified. Their work is very important and should not be interrupted, due to the fact that the load on them can be at any time. But there is another side, this is the time of their active loading. The percentage of unused resource - fluctuates around 90%. That's how I got the idea to do vertical scaling with auto-scaling and auto-scaling to a cheaper configuration. If you can suggest or give an idea for implementation, I will be very grateful to you.