perf(runner): VM startup optimization plan - reduce from 20-25s to 3-5s

[lancy]

Overview

This is the tracking issue for the Firecracker VM startup optimization effort. The goal is to reduce VM startup time from 20-25 seconds to 3-5 seconds.

Current State

Phase	Time	Bottleneck Level
Rootfs copy (2GB)	~12s	🔴 Critical
Network setup	~0.5s	🟢 Low
Firecracker spawn	~0.2s	🟢 Low
VM boot (kernel + systemd)	~5-8s	🟡 Medium
SSH readiness	~3-5s	🟡 Medium
Post-boot setup	~1-2s	🟢 Low
Total	20-25s

Target State

Phase	Current	Target	Optimization
Rootfs copy	12s	0.003s	BTRFS reflink
VM boot	5-8s	1-2s	Minimal init
SSH readiness	3-5s	<1s	Vsock
Total	20-25s	3-5s

Optimization Tasks

Phase 0: Foundation (Dev Mode)

• feat(runner): add benchmark command for VM performance testing #1165 - Dev mode for local testing & benchmarking

Phase 1: High Impact (~12s savings)

• perf(runner): implement BTRFS reflink for instant rootfs copy #1162 - BTRFS reflink for instant rootfs copy

Phase 2: Boot Optimization (~3-5s savings)

• perf(runner): implement vsock for host-guest communication #1163 - Vsock for host-guest communication
• perf(runner): replace systemd with minimal init for faster boot #1164 - Minimal init to replace systemd

POC Validation Results

Tested on target machine (35.87.132.135 - ARM64, 16 CPU, 31GB RAM, gp3):

Test	Result
gp3 write speed	174 MB/s
EXT4 reflink	❌ Not supported
BTRFS reflink (2GB)	✅ 0.003s
RAM disk speed	1.1 GB/s
KVM support	✅ Available

Success Metrics

Metric	Current	Target	Status
VM startup time	20-25s	<5s	🔄 In Progress
Rootfs copy time	12s	<0.1s	🔄 Pending
SSH ready time	3-5s	<1s	🔄 Pending

Recommended Order

1. feat(runner): add benchmark command for VM performance testing #1165 Dev Mode - Enables testing and benchmarking for other optimizations
2. perf(runner): implement BTRFS reflink for instant rootfs copy #1162 BTRFS - Highest impact, well-validated approach
3. perf(runner): replace systemd with minimal init for faster boot #1164 Minimal Init - Reduces boot overhead
4. perf(runner): implement vsock for host-guest communication #1163 Vsock - Most complex, do last

References

• Research document: /tmp/deep-dive/runner-perf-optimization/research.md
• Innovation document: /tmp/deep-dive/runner-perf-optimization/innovate.md