Generalizability of the Observed Performance Crossover for Node.js under Memory Constraints

Ascertain whether the performance crossover in Node.js workloads under severe memory constraints—where Docker containers maintain stable performance while Nanos unikernel performance degrades sharply—is a generalizable phenomenon across other unikernel implementations such as Unikraft or MirageOS, or is specifically caused by the Nanos unikernel’s memory management strategy.

Background

The paper identifies a performance crossover for JIT workloads (Node.js) under tightening memory constraints: below a certain threshold (notably around 128 MB), Docker’s performance remains stable while Nanos degrades sharply in throughput and latency.

In the conclusion, the authors explicitly state uncertainty about whether this crossover is a general phenomenon of unikernels or specific to the Nanos unikernel’s memory management. They suggest comparative evaluations with other unikernel systems (e.g., Unikraft, MirageOS) to resolve this uncertainty.