Ceph and NVMe/TCP
summary
NVMe/TCP is in Linux kernel 5.0 and delivers NVMe-over-fabric without changing network gear; integration with Ceph moves metadata off local disks so clusters avoid freeing slots to add NVMe. Implementation from Lightbits Labs claims lower CPU load and runs workloads like RocksDB and Apache Spark with tens of microseconds extra latency. Appliance and software-defined targets expose NVMe/TCP for data and metadata, support replication across nodes without performance loss, and plan a key-value interface to replace RocksDB inside the OSD.
AI-written summary · about
related talks
39mSuper Fast Real-time Data Processing on Cloud-Native Architecture
24mBreaking the TCP Barrier: Accelerated I/O for S3 With RDMA
34mEvaluating the Performance and Power Efficiency of Ceph Storage Configurations
43mComparing Performance of NVMe Hard Drives in KVM, Baremetal, and Docker
36mUsing NGINX as a CDN for Ceph Radosgw
29mFragmentation in Ceph: Why It Happens and How We Fix It
40mAchieving the Ultimate Performance with KVM
19mAsynchronous I/O Stack: A Low-latency Kernel I/O Stack for Ultra-Low Latency SSDs