MemSweeper: Virtualizing Cluster Memory Management for High Memory Utilization
summary
MemSweeper virtualizes a shared memcached heap so pages move across tenants and slab classes based on measured reuse rather than static partitions, implemented as a tracker plus per-tenant shadow queues and an AVL tree reuse-distance method to get O(log n) misses (avoiding the linear slowdown that made tenants unresponsive). It computes marginal utility per-page and runs MaxSweeper (maximize system hit rate) or FairSweeper (limit victim loss), evaluated on memcached with CloudSuite; MaxSweeper raises overall hit rate but FairSweeper prevents aggressive tenants from starving others.
AI-written summary · about
chapters (4)
related talks
20mUniMem: Redesigning Disaggregated Memory within A Unified Local-Remote Memory...
18mHermit: Low-Latency, High-Throughput, and Transparent Remote Memory...
16mCacheSack: Admission Optimization for Google Datacenter Flash Caches
19mFineMem: Breaking the Allocation Overhead vs. Memory Waste Dilemma in Fine-Grained...
45mBeyond Machine Boundaries: Experience with Memory Disaggregation
16mDisaggregation and the Application
17mDuhu: Shared Disaggregated Memory for Distributed Data Processing Frameworks
16mA Tale of Two Paths: Toward a Hybrid Data Plane for Efficient Far-Memory Applications
