ECS or EKS?

Both mistakes come from deciding by reputation. This lesson is about deciding with evidence: what each platform really costs this team, measured, and written down so the decision survives someone arguing the other side.

ECS and EKS do the same job. Both take your image, keep a set number of copies running, replace them when they fail, release new versions gradually, and put a load balancer in front. They run the same container image, in the same VPC, behind the same kind of load balancer, with the same IAM and CloudWatch.

The difference is how much of the platform you run yourself, and what you get in return.

The comparison, on the five things that matter

Prices are for us-east-1 at the time of writing, and they change. Check the ECS and EKS pricing pages before you quote numbers.

What you actually saw, running the same service on both

These come from the ECS on Fargate and Kubernetes and EKS labs, where the same image ran on both:

Six questions that decide it

A recommendation isn’t a feature list. It answers: for this team, running this app, what should they do?

When each one fits

Answer each question for your own team, or fill in QuickCart’s answers. Then change one answer at a time and watch which ones move the recommendation.

What to notice: no single feature decides it. The answers that move the recommendation most are about people: who runs it, what they know, and what already runs. That’s why the memo you’ll write leads with the team, not the technology.

Free — you write a document. Nothing is created in AWS.

You’ll write a one-page recommendation, the kind a team lead or architect asks for before a platform decision. About 30 minutes. It’s best done after the ECS and EKS labs, so you can use your own numbers. If you haven’t done them, use the numbers in the idea.

1. Collect the evidence

   Make a table of facts before you write any opinions. For each row, use your own measurement if you have one. Otherwise use the course’s figures and say so.

   What	ECS on Fargate	EKS	Where it comes from
   Time to create the cluster	seconds	your time, or about 15 to 20 minutes	your lab notes
   Steps to connect a load balancer	one form section	your count of commands	your lab notes
   Failed requests during a release under load	your number	your number	Releases without downtime
   Undoing a bad release	automatic	one command	the labs
   Cost before any app runs	$0	about $73 a month	AWS pricing pages, checked today
   Servers to look after	none	your nodes, unless on Fargate	the labs

   You should see

   A table where every number has a source. If a cell says “I think”, find the number or leave the row out.

2. Answer the six questions for your team

   Use the decision helper in See it. Answer honestly for the team that would run the platform, not for an imaginary bigger one.

   Write down the recommendation it gives, and which answers pushed it hardest. Those answers are the reasons you’ll lead with.

3. Write the memo

   Create a file called recommendation.md and fill in this template. Replace every line in capitals. Keep it to one page.

   recommendation.mdCopy

   Recommendation: ECS OR EKS for new services.
   
   Why: THE ONE DECIDING REASON, IN ONE SENTENCE.
   
   What we measured:
     Time to create a cluster:        ECS NUMBER   EKS NUMBER
     Steps to connect a load balancer: ECS NUMBER   EKS NUMBER
     Failed requests during a release: ECS NUMBER   EKS NUMBER
     Undoing a bad release:           ECS ANSWER   EKS ANSWER
     Cost before any app runs:        ECS NUMBER   EKS NUMBER
   
   Operational burden: WHAT EACH ONE MAKES US LOOK AFTER, AND WHO WOULD DO IT.
   
   Cost at our scale: THE FIXED COST, AND PEOPLE'S TIME, COMPARED WITH OUR BILL.
   
   Team readiness: WHO KNOWS WHAT TODAY, AND HOW LONG LEARNING WOULD TAKE.
   
   Migration risk: WHAT ALREADY RUNS, AND WHAT RUNNING TWO PLATFORMS WOULD COST.
   
   The strongest argument against this, and our answer:
     ARGUMENT.
     ANSWER.
   
   What would change our mind: THE FACTS THAT WOULD FLIP THIS DECISION.

   Section	What goes in it
   Recommendation	The answer first. A reader who stops here still knows what you think
   Why	One reason. If you need three, the first one isn’t strong enough
   What we measured	The evidence table, short
   The four middle sections	One short paragraph each, with a number where you have one
   The strongest argument against this	The best case for the other side, stated fairly, then your answer
   What would change our mind	The facts that would flip the decision. This is what makes it a judgement, not a preference

   You should see

   One page. The first line is the answer. Every paragraph has at least one fact or number.

4. Argue the other side

   Before anyone else does, write down the strongest argument against your recommendation, and answer it. These are the ones you’ll hear:

   If you recommend ECS, expect	If you recommend EKS, expect
   “Kubernetes is the standard. It’s easier to hire for.”	“Now we run two platforms. Who looks after both?”
   “EKS stops us being locked in to AWS.”	“Is leaving AWS actually planned, with a date?”
   “Kubernetes has tools ECS doesn’t.”	“Which tool does this app actually need?”
   “We’ll outgrow ECS.”	“What will we be doing in the months it takes to learn this?”

   Pick the one that worries you most, and put it in the memo with your answer.

5. Check it

   Read the memo against this list, and fix anything that fails:

   Check	Passes when
   The answer comes first	The first line says ECS or EKS
   Every claim has evidence	Each paragraph has a number, a measurement or a source
   Prices are dated	Every price says where and when you checked it
   The other side is fair	Someone who disagrees would recognise their best argument
   It can change	“What would change our mind” names specific facts
   It fits on one page	If not, cut adjectives first, then repeated points

6. Clean-up step: Clean up

   Nothing was created in AWS in this lab, so there’s nothing to delete. Keep recommendation.md; Project 3 asks for exactly this memo, with your own measurements.

   If you still have the ECS service or the EKS cluster from the earlier labs, delete them now using those lessons’ clean-up steps. An EKS cluster costs money every hour it exists, even when idle.