Iacovos Kolokasis from FORTH presented their work-in-progress paper “SmartSweep” at MPLR 2025!
Big data frameworks (e.g., Spark, Neo4j) often extend the JVM heap into remote memory, but GC in this setting is a nightmare: high network traffic, delayed reclamation, and nasty OOM errors.
💡 SmartSweep tackles this head-on:
⚡ Dual-heap design with approximate liveness info
🗑️ Selective reclamation of garbage-heavy regions — no scanning, no compaction
📉 Cuts remote memory usage by up to 49%
✅ Matches TeraHeap’s performance without the OOM risks
Early results are promising — and I’m excited to discuss this at MPLR’25. Stay tuned!
Preprint: https://zenodo.org/records/17213496
