Published release 1.5.0 of the Vitastor data storage system with support for clustered file system (VitastorFS). Vitastor is a distributed block data storage system, which serves as storage for virtual machine images or container disks, developed by the author since 2019. Starting with release 1.5.0, it also functions as a clustered file system. VitastorFS is mounted via the NFS 3.0 protocol from either locally or remotely hosted server(s).
Vitastor is primarily aimed at SSD and SSD+HDD clusters (but it also works in configurations with only hard drives) with at least a 10 Gbps network, supporting TCP and RDMA. With good hardware, it can achieve 4KB read and write latencies around ~0.1 ms, which is approximately 10 times faster than most popular SDS and in-house SDSs of major cloud providers — both Ceph and typical cloud SDS yield write latencies of 1-2 ms even on very fast disks.
For example, in a recent test with Samsung PM9A3 NVMe disks and Mellanox ConnectX-4 network cards with RDMA, the ideal write latency was > 0.1 ms when using replication and 0.1 ms for reading / 0.2 ms for writing when using error correction codes.
Vitastor supports QEMU driver, NBD and NFS protocols, Proxmox drivers, Kubernetes CSI, OpenStack. Additionally, an experimental Kubernetes operator and a Prometheus exporter are available for Vitastor.
Vitastor server components are distributed under its own 'network copyleft' license, VNPL (Vitastor Network Public License). The idea of VNPL is to extend the copyleft's applicability not just to modules explicitly linked with the code, but also to services that interact with Vitastor over the network, specifically created for it (for example, any wrappers for internal use in commercial cloud). Otherwise, the text of VNPL is copied from GNU GPL 3.0.
Source: opennet.ru
