Skip to content

Machines & Storage

The platform’s launcher deployed VMs in three steps:

  1. Select a region, name the instance, and choose a flavor (CPU/RAM/disk profile).
  2. Choose the VM image version — or an app source such as a Git URL.
  3. Provide security credentials and launch.

Users could connect to running machines over SSH/console, attach and detach disks, and manage lifecycle from the web UI or API. Docker Machine integration was also supported for driving Docker hosts.

Block storage volumes expanded VM datastores and could be detached and re-attached to another VM — a low-bandwidth way to move large datasets between servers. Because a volume is a separate partition from the OS partition, volumes could be removed later without affecting the server — safe scaling by design.

Storage nodes doubled as object storage: clients could store any data type (images, video, arbitrary files) directly via API or through the UI. Object storage was configured redundantly — multiple copies of client data, with optional multi-geographical placement for increased redundancy.

VM state could be snapshotted as a backup to revert to later, or as a scaling tool — snapshot a VM, delete it, and relaunch the snapshot when needed again. Snapshots were stored automatically in the object storage system and billed accordingly.

The platform’s HA design stored a secondary copy of VMs through a network RAID-1-style system, avoiding expensive SAN hardware. Two reference setups:

NodeStorageRole
Compute & Storage Node 12×HDD/SSDOne drive for VMs, one storing node 2’s data
Compute & Storage Node 22×HDD/SSDMirror arrangement of node 1
Failover node1×HDD/SSDStandby; on failure, rebuilds from the surviving copy (~1 per 10 compute nodes)

Double-copy mode stored two copies of each compute node at the cost of doubled storage capacity.

NodeNetworkRole
Compute nodes1 Gbps privateAll drives usable by VMs; copies pushed to storage nodes
Storage nodes10 Gbps privateCluster storing copies from many compute nodes
Failover node1 Gbps privateStandby; pulls backups from storage and reboots VMs automatically

On a hardware failure, VMs migrated to a new host with no data loss and minimal interruption — the property that qualified the platform as a true cloud rather than static VPS hosting.