Bash spike to measure EC2 deployment speed. Goal: learn where time goes, apply to brownfield backend.

The ~16s Pending→Running phase is fixed AWS infrastructure overhead. It cannot be reduced by:

• Instance type selection (680ms variance across t3→r7i)
• Region selection (2.1s variance, mostly network latency)
• Stripping cloud-init (made it slower!)
• Hibernate vs stop/start (only 0.5s difference)

The only way to achieve <10s startup is warm pools with running instances.

• Cold Baseline - Where does the time go?
• Start-Stopped - Pre-warmed EBS/ENI
• Hibernate - Resume from hibernation
• Minimal AMI - Stripped cloud-init (don't do this)
• Warm Pool (SSH) - Theoretical floor
• Warm Pool (EIC) - Production recommended
• Instance Type Sweep - t3→r7i comparison
• Region Sweep - Cross-region comparison

# 1. Setup AWS resources
./scripts/setup-security-group.sh
./scripts/setup-keypair.sh

# 2. Run baseline benchmark
./bench/cold.sh ~/.ssh/instant-env-admin.pem

# 3. Run warm pool benchmark (production approach)
./bench/warm-pool-eic.sh

# 4. Cleanup when done
./scripts/cleanup-all.sh

T0  Script starts
 │
 ├─ aws ec2 run-instances
 │
T1  Instance ID returned (~5s)
 │
 ├─ aws ec2 wait instance-running
 │
T2  State = "running" (~18s) ← THE BOTTLENECK
 │
 ├─ nc -z (TCP probe)
 │
T3  TCP port 22 open
 │
 ├─ ssh (auth handshake)
 │
T4  SSH auth success (~3s)

Cold Total = ~26s
Warm Pool = ~4s (skips T0→T2)

For production: Use EC2 Instance Connect with warm pools

• 3.9s access time (85% faster than cold)
• No pre-existing SSH access needed
• Ephemeral keys (60s validity)
• IAM-integrated, auditable