The Small Container: Lightweight Compute for Multi-architecture Cloud-Edge Systems

Docker and WebAssembly (Wasm) are playing increasingly important roles in
modern software development, each offering unique benefits in portability, per formance, and deployment efficiency. As Wasm evolves beyond the browser into
cloud-native and edge computing environments, its integration with container
runtimes prompts a closer examination of performance trade-offs compared to
traditional multi-platform containers.
In contrast to conventional Docker containers, Wasm binaries offer a portable,
compact, and secure deployment format. In this paper, we extend our
prior investigation by introducing two new benchmark suites: benchx and
sqlite-multiarch to explore deeper performance characteristics across Wasm
and native multi-architecture containers using containerd, focusing on four plat forms: AMD64, ARM64 (Nvidia’s Jetson Nano and Orin), and RISCV64 (StarFive
VisionFive2). We analyze pull time, startup latency, and image size across both
native and Wasm runtimes.
Our results show that Wasm containers reduce image size by up to 70% and
achieve up to 25% faster cold pull times compared to native containers. With the
sqlite-multiarch benchmark, we observe Wasm startup overheads compared
to native execution, which highlights both performance gaps and opportunities
for runtime optimizations. Across all platforms, Wasm containers executed via
containerd consistently outperformed Docker-based setups.
We demonstrate that Wasm runtimes can effectively support realistic data intensive workloads such as full-text search, JSON parsing, and R-tree spatial
queries, while offering improved deployment efficiency and better performance
isolation. These findings reaffirm Wasm’s potential as a complementary compute
layer for heterogeneous, multi-architecture cloud-edge deployments.