Coming Soon

Russel

Serverless Runtime Environment

Technical blogs soon. Follow on X for updates.

Exploring structural deduplication for microVMs.

~1MB
payload
247ms
deploy
russeld — deploy -p 3000:3000 --vm-id test-vm ./go-dashboard
$ russel deploy -p 3000:3000 --vm-id test-vm ./go-dashboard
→ Initializing microVM runtime...
→ Loading foundation layer (Go 1.22)...
→ Mapping code payload (1.2MB)...
✓ Deployed in 247ms
$ curl -I localhost:3000
HTTP/2 200 — 12ms TTFB
Benchmark Data

Performance Results

Comparing cold starts and build times between Russel and legacy container runtimes.

Cold Build

185ms vs Docker: 11.4s

62x faster

Incremental Build

242ms vs Docker: 955ms

3.9x faster

Improving Soon

VM Boot Time

11.8s vs Docker: 1.7s

Currently slower than containers.
MicroVM snapshots arriving in Phase 1.

Disk Scaling

~1MB

Per new instance via QCOW2

Cross-App Dedupe

304ms

New app with shared runtime

Beyond Container Layers

Traditional container caching works at the image layer level. Russel redefines deduplication by moving it into the host virtualization stack.

Instead of layering filesystems that must be pulled and unpacked, we map immutable structural blocks directly into isolated microVMs as read-only volumes. This eliminates the "cold pull" bottleneck entirely.

  • Block vs Layer: Deduplication happens across all tenants at the MicroVM transport level.
  • Zero-Copy: Foundations are shared once in host RAM and mapped explicitly via virtio-fs.
  • Atomic Deploys: App updates are just pointer swaps in the shared storage engine.
The Core Concept

Exploring serverless efficiency

Investigating if we can bend the trade-off between performance and efficiency.

The Problem

Containers bundle entire runtimes. Every deploy duplicates 100MB+ blocks.

The Solution

Separate code from shared foundations. Map payloads to pre-warmed VMs.

Isolation

Hardware-isolated kernel per tenant. Zero-trust by design.

Updates

Patch foundation once, apply to all VMs. Security without rebuilds.

System Architecture

How Russel Works

Modular architecture: CLI → Control Plane → Rust Engine + Shared Cache

CLI Tool Go
Control Plane (russel-d)
gRPC
Engine Layer

russel-engine (Rust)

VM Lifecycle
Networking
Cloud Hypervisor (KVM)
Cache Layer

Shared Runtime Cache

Immutable, content-addressed, and shared across all VMs.

/shared/runtime-cache
Isolated MicroVM
Guest Kernel vmlinux
Read-Only Cache Mount virtio-fs
Vision & Roadmap

Future Directions

Phase 1

Speed

Implementing faster boot using VM snapshots to reduce startup time from seconds to milliseconds.

Phase 2

Scaling

Adding horizontal scaling with built-in reverse proxy for fast load balancing and traffic management.

Phase 3

Orchestration

Expanding to multi-node orchestration for distributed workloads and maximum high availability.

Long-term

Vision

Hardware virtualization meets declarative state management for the next generation of cloud infrastructure.

Follow the progress

Technical blogs soon. Follow on X for updates.