7 · Compute Choices — Revise¶

Performance · 24% · Compressed recall. Read the Learn guide first.

Pick the compute¶

Long-running (>15 min) / OS control / steady high use?  → EC2
Short, event-driven, ≤15 min, scale to zero?            → Lambda
Containers, no servers to manage?                       → Fargate (ECS or EKS)
Containers + need node control / max density?           → ECS/EKS on EC2
Standardized on Kubernetes?                             → EKS
Large-scale batch?                                      → AWS Batch

EC2 family → bottleneck¶

Family	Optimized for
T / M	General purpose (T = burstable)
C	Compute (CPU-bound, batch, HPC)
R / X	Memory (in-memory DB, big cache)
I / D	Storage (high local disk IOPS/throughput)
P / G	GPU (ML, graphics)

Lambda limits¶

Limit	Value
Max timeout	15 min (900 s)
Memory	128 MB – 10 GB
`/tmp`	512 MB – 10 GB
Package	50 MB zip / 250 MB unzip / 10 GB image
Default concurrency	1,000 / Region (soft)

Containers¶

	ECS	EKS
Orchestrator	AWS-native	Kubernetes
Launch types	Fargate (serverless) or EC2 (you manage nodes)	same

Flashcards¶

Job runs 40 minutes, want no servers. Lambda?

No — Lambda caps at 15 min. Use Fargate (or EC2/Batch).

"Serverless containers" vs "serverless functions"?

Serverless containers = Fargate. Serverless functions = Lambda.

Need Kubernetes without managing the control plane?

Amazon EKS (with Fargate to also drop node management).

In-memory database needs lots of RAM — which EC2 family?

Memory optimized: R or X.

Latency-sensitive Lambda hit by cold starts. Fix?

Provisioned concurrency.

Steady, always-on, high-utilization compute — cheapest?

EC2 with a Savings Plan / Reserved Instances (Topic 11).

Common traps¶

Lambda = 15-minute cap — the top rule-out; longer ⇒ Fargate/EC2/Batch.
Fargate = serverless containers; Lambda = serverless functions.
EKS = Kubernetes; ECS = AWS-native — both run on Fargate or EC2.
Match EC2 family to bottleneck (R/X memory, C CPU, P/G GPU, T burstable).
Spiky/event-driven → serverless; steady high use → EC2 + commitment.