AWS EC2 Instance Types to use?

Hi,

I am preparing to operate a stake pool and I want to reserve instances in AWS to save costs but I need to make sure the types are correct and also future proofed.

Is anyone running in AWS? What instance types are you using?

I read this thread → Amazon AWS (pool stake operators) Whats the best / cheapest setup?

@ kaverne is suggesting a r6g.medium (1 core 8gb ram)

Before reading that thread I was thinking of using their burstable instances (t3.large - 2 core 8gb ram).

Any thoughts?

contabo vps ; plan M (or even L) if u are looking to minimize the cost (the cheapest services )

1 Like

Thanks but my last few dev projects were on AWS so its fresh in the mind :slightly_smiling_face:

Besides if @kaverne is correct 3 reserved r6g.medium instances for 3 years paid up front is about $1500, which works out to be around $14\month\server which isn’t too bad

1 Like

Hi @Cardano-Cal ,

r6g works well, but I wouldn’t commit to 3 years plans as specs might need a change eventually (cpu, mem, better instance etc …).
At the moment I got a 1 year plan, I can’t remember the exact maths, but it doesn’t matter anyway. The rough idea is it’s around 40% rebate. Which means that 7 months in your plan, it’s like all free for the remaining 5 months compared to on demand price.
That makes the commitment absolutely acceptable as the instances are really good.

I’m an AWS, linux, devops and solution architect in my daily job too so I guess that makes it easier for us.
I try to not enter into debate as everyone seems to be very defensive with their own solution but you can’t really compare small cloud like offering and AWS (or Azure).
If you design your architecture well you can really benefit from all their infrastructure and tools for near free. And that makes a lot of capabilities to make really cool and fun stuff.

I’m currently contemplating the idea of using another instance type for the block producer (still in the Graviton2 type).
But I validate that I’m using r6g right now and very happy with it.

Dont whatever you do sign up for Amazon AWS reserved instances. You will be locked into a year and be charged every month even if you dont use the servers. (though you can sell on the marketplace mine have been on the market 2 months and no one has taken off my hands, they said they would cancel it which i hoping because its $250 a month till may for servers im not using!) I dont recommend partial upfront or locking into a year unless your 100% sure this is the route you will go. Its much cheaper building 2 pcs at home. (check my server build, a bit more than what is needed: Computer parts, laptops, electronics, and more - Newegg United States $1,700 for both servers but total is around $1900 shipped / taxes)

The cheapest servers on Amazon are the Gravitron processors t4gs, etc. which isnt officially supported by Cardano even though there is a lot of compiled binaries available, since many people use ARM/ARCH based Raspberry Pis.

I moved to my home away from Amazon AWS even though i have a bunch of cheap little $6 servers running my various websites

I would LOVE to run it at home, my PCs would eat these nodes up lol but unfortunately the infrastructure in my area is not stable, too often our internet is down due to maintenance or our power is out.

So cloud providers are my only option and on-demand instances are just crazy expensive, it would cost the same for 1 server on-demand as it does for 3 servers reserved.

Just out of curiosity, what did you reserve that costs you $250\month?

Hey kaverne,

I was hoping you would reply :slight_smile:

Im between reserving for 1 year and 3 years, 1 year is about half the price of 3 years so there is a big saving going over 3 years.

I totally agree, for now I will just run the nodes but later I could leverage off other AWS services to really provide a rich experience.

Thanks, this is what I was looking for :+1:t3:

Id be interested to hear how you have built your environment, I am also using CloudFormation and looking to automate the release process using CodePipeline, mine would work something like this:

  • pull latest branch of cardano
  • push to CodeCommit
  • trigger starts CodePipeline
  • CodeBuild will run all required commands and build the project
  • CodeDeploy to push those changes to TestNet
  • I can then check all is 100% on TestNet
  • Manual approval
  • CodeDeploy to push changes to MainNet

8 gb of ram is not enough. It used to be, but not any longer.

I would suggest using a t3.xlarge, which gives you 4 cores and 16 gb of ram. The 16 gb of ram will give room for growth over the next year, and the 4 cores will help with missed slots during epoch changes.

Reserve an instance for a year, not three, because, as someone else pointed out earlier, requirements could change over the course of three years but less likely during the next year.

1 Like

Thanks for the advice.

Have you had issues running it with 8Gb of ram?

I asked my credit card and it said no :joy: 3 t3.xlarge reserved for a year is $2500, the r6g.medium is $750, so there is a big difference in cost

:+1:t3:

Just use Contabo if you want to save money vs AWS I dont recommend running a Cardano node unless you use a Gravitron processor which isnt officially supported.

I am using a r6gd.large and they perform quite well - especially when starting the node with the parameters defined by SPIDR ( weebl2000 )
Just note they are AARCH64 machines so many common tools like CNTools are not compatible with these CPU’s. So if you were to copy the cardano-cli from this instance for an air-gapped machine be aware you would want to put it onto a similar AARCH64 machine I learnt that the hard way…
I am having some performance issues with contabo instances at various times in the day also.

I am comparing these to Linode currently as I shop around to see what gets the best reliability/performance/cost.