memory-tiering

Tiered Memory Management Beyond Hotness (OSDI'25)

Hardware-accelerated Linux kernel block driver and multi-tier memory cascade (ZRAM ➔ PCIe VRAM ➔ NVMe). Opportunistically leases idle GPU memory as a zero-allocation, revocable swap cache with sub-millisecond latency, zero SSD wear, and crash-safety on Linux & WSL2.

A tiered-memory system design for workloads that don't fit in RAM: measure the working set, pin the hot tier, stream the cold tier from flash. Ships the residency calculator, measurement harnesses, and the build recipes behind it. Predictions validated against public benchmarks.

MegaMmap: Blurring the Boundary Between Memory and Storage for Data-Intensive HPC Workloads. A software distributed shared memory system that enables infinite memory capacity through intelligent tiered DRAM and storage management.

Recipe to run a memtier benchmark on a cluster of KVM-hosted Redis servers

VibeDrift - Run any LLM on your own hardware. Bypass the VRAM wall with CPU/RAM inference, MOE expert offloading, and 4-bit quantization. No Cloud, no Subscription.

Reference Linux control plane, RTL, and FPGA emulation scaffold for KV-CPU semantic KV-cache orchestration. Patent pending in India (App No. 202641056309).

Experimental Linux RFC for an HBF/CXL-era AI memory control plane: runtime hints, prefetch, placement, and tiering.

Deadline-aware KV-cache scheduling for protecting decode-critical request-state under long-context LLM inference pressure.

A systems research platform for semantic KV-cache orchestration, topology-aware memory placement, distributed prefix reuse, and rack-scale inference memory simulation.

Profile Qwen3.8 n-gram table access heat and offload the cold majority to SSD

Reference Linux kernel interface for workload-aware memory hints in AI systems — /dev/mem_hint maps LLM phases like prefill, decode, agentic execution, and training to memory-policy signaling paths.

QEMU CXL type-3 memory + the Linux tiering path, re-examined under a calibrated DDR5/CXL timing model. Advertised 1.62x, emulated 1.01x, modelled 1.90x - and what that gap does to placement policy. Includes a conference-style paper.

Software proof-of-concept for entropy-based memory tiering on Linux.

Run local LLMs on your own hardware with this C++ powered inference engine.