Enabling Big Memory workloads to run
summary
MemVerge’s software-defined memory layers above Linux, auto-tiers pages between DRAM and Intel Optane persistent memory, and provides copy-on-write memory snapshots that enable fast checkpoint/restart, migration and parallel cloning. Benchmarks show 300M Redis keys serialized in 15 minutes vs dumped to PMEM in 0.5s, a single-cell RNA pipeline running 60% faster, and a metagenomics job that normally runs 11 days completing on one commodity Ice Lake box by using ~2 TB of memory. Integration targets OpenStack/KVM, Kubernetes operator and Loki.
AI-written summary · about
chapters (3)
related talks
25mGreenOps at Scale: The 50TB Memory Diet for a Petabyte Platform
14mPD3: Prefetching Data with DPUs for Disaggregated Memory
33mMemory Forensics Using Virtual Machine Introspection for Cloud Computing
01h 07mOpenCAPI BoF - A Memory Centric Fabric for a Data Centric World
22mGraviton
24mCloud Workloads At Petabyte Scale
25mOptimize productivity by meeting next-level demands of the AI era
26mCrafting a Sustainable Datasphere: Storage Innovation