Joviam’s distributed block storage system (DSAN Storage) combines commodity cloud hardware with high speed Infiniband to create a reliable, scalable storage alternative to traditional SANs and direct attached storage.
Reduced flexibility and redundancy are common issues found when using direct attached storage. Joviam overcomes these issues through a virtual SAN (DSAN) across our cluster, using a distributed storage architecture over supercomputer technology (Infiniband). This architecture provides greater cost efficiencies, reduces latency and overcomes the points of failure typically found with Centralised SANs.
Joviam’s hyperconverged architecture means that all nodes within our cluster have compute and storage resources, allowing horizontally scaling of our platform on all tiers. Storage assets on each node are virtualised and aggregated into our distributed SAN (DSAN).
Joviam utilises Infiniband technology, unlike traditional cloud providers that use Ethernet. Commonly found in supercomputers (read: Large Hadron Collider), Infiniband allows us to give our users significantly higher IO performance, all the while maintaining data consistency through synchronous replication.
Compared to Inifiniband, Ethernet is highly latent and often bottlenecks distributed systems on other Ethernet-based cloud platforms. In contrast, Infiniband significantly reduces the latency of reading and writing data to remote servers in a cluster, through a combination of its architecture and allowing the use of RDMA for storage transport. As a result, the number of IOPs possible across a cluster is significantly increased.
Whenever a Virtual block device is created in Joviam’s DSAN, the device’s backing blocks are distributed and replicated across multiple nodes in our cluster (over dual redundant 40Gbit QDR IB).
This means that, for every write to disk:
Node availability within the cluster is constantly monitored by our SDS system to ensure prescribed levels of data replication are consistently maintained.
If our SDS detects that a node is either having issues or is offline, data replication of the missing blocks to other nodes within the cluster begins immediately, thereby maintaining prescribed levels of data replication. When the original node is recovered, any changes made to storage are automatically updated and replicated – this entire process of replicating storage after an event is entirely automatic, requiring no user involvement.
Joviam utilises a hyper-converged architecture. This means that storage and compute are present on each server node, but a virtual machine does not have to be booted on the same node as where its storage is located.
The key feature of this architecture is flexibility. With Joviam you do not have to buy a fixed amount of storage with each VM and are only billed for the resources that are currently provisioned (and therefore not available to other parties). For example, if you turn off your VM we will stop billing you for CPU and memory, and you will only be billed for storage/IP that is still provisioned. This is in contrast to most other cloud providers that require you to buy a fixed amount of resources and continue to bill you for all resources after a machine is switched off.
As all our storage is network backed, we can mount any block device on any hypervisor. This means we can maintain significant flexibility when migrating disks between virtual machines. It also provides failover recovery, so that if a node fails, the virtual machine can be provisioned on a different node whilst maintaining access to the same storage. This also adds the ability to live migrate virtual machines around the cluster and perform maintenance without downtime.
Joviam’s SDS platform enables online snapshotting and cloning of block devices, meaning that disks currently in use are able to be cloned. The clone is performed at the block layer almost instantly and can be immediately mounted. This allows for ‘point-in-time’ snapshots or rapid scaling out of multiple machines from a clone template.