Hey everyone! We're a group of undergraduate engineering students working on our final year project, which involves hosting our AI backend on AWS. For our evaluation, we need to manage about 25 users at the same time to demonstrate how scalable our application is. We're wondering if it's feasible to create around 15 g5.xlarge EC2 instances on a fairly new AWS account for about 5 to 8 hours. Are there any restrictions we should be aware of for this account, and if so, what steps do we need to take to utilize that many instances, like adjusting service quotas? If anyone has dealt with this before, we'd really appreciate your insights on tackling these challenges!
4 Answers
Unfortunately, it's likely that you won't be able to create 15 g5.xlarge instances right off the bat, especially since you're using a new AWS account. Most new accounts start with no quota for GPU instances. I recommend checking your vCPU quota in your AWS dashboard under Service Quotas and looking at the on-demand G series instance types. If you find that you have a quota of zero, you’ll need to submit a request to increase your quota for the instances you plan to use. This process can take some time, so it’s best to get started on that as soon as possible. Keep in mind that requests could go through human review, so be ready to explain your needs for the increase!
Are you sure you need 15 instances? That seems like a lot. Have you tried testing your app's load with, say, a couple of instances first? Sometimes just optimizing the usage can save you money and resources. Also, make sure you account for the costs; running those instances can get pretty pricey!
I've tested the app with one instance and it managed 3 users, but we want to have the flexibility to scale up, hence the number.
Just a heads up, running those g5.xlarge instances will set you back about $15/hour. You need to weigh whether your project is worth that investment and if you're prepared to handle those costs.
I've got the budget covered and can reimburse the charges if needed.
Have you thought about why you actually need that many instances? Sometimes, instead of spinning up more hosts, you can achieve greater performance through process-level optimization. What's your workload plan, and have you tried other scaling methods?
I submitted a quota increase request for G instances not long ago and it got approved in about 30 minutes! Do you think the 10 days left on my request will be enough?