Distributed Pseudo-Filesystem for VM Block Storage
VM Cluster Filesystem
VMFS-class cluster filesystem — 32-node clusters, 14000 VMs, thin volumes, snapshots and live migration.
2024Shipped
5engineers
3Screens
ClientSector: Enterprise virtualization vendors · NDA
Overview
A cluster pseudo-filesystem for virtualization workloads, designed as a VMFS-class equivalent for shared block storage. Uses a vector-clock-based DLM written through the underlying disk on storage that does not expose SCSI reservations; falls back to SCSI reservations or ATS (hardware-assisted locking) primitives where the array supports them; uses Ceph's native locking when running on Ceph. Backends: LVM block storage or Ceph. Supports thin volumes, snapshots and live VM migration. Production-validated at 32 nodes per cluster, 14000 VMs and 500 concurrent disk expansions.
What ships with it
- → Vector-clock-based DLM through the underlying disk — works without SCSI reservations
- → Falls back to SCSI reservations or ATS where the storage supports them
- → Native Ceph locking when running on Ceph
- → Thin provisioning, snapshots and live VM migration
- → Validated at 32 nodes, 14000 VMs and 500 concurrent disk expansions
Screens
01/03
Click any screenshot to open full-size view
Numbers that matter
32Nodes per cluster · validated for
14000VMs · validated for
500 parallelConcurrent disk expansions · validated
LVM + CephBlock backends
Stack
GoRustCLVMCephLinuxSCSI
Want one of these for your team?