- Designed for optimal distributed replication latency & two layers of high availability:
- - Data Deltas replicated peer-to-peer between Agents for optimal latency & reduced central load
- - CRDT algorithms enable Agents to perform writes while temporarily offline
- - Seamless Agent failover, even on multiple simultaneous data-center failures
- - HighlyAvailable/Elasitic on two levels: Cluster level & Mesh of Clusters level
Replication is peer to peer between Agents, this makes for a highly robust system, yields optimal latency for replication between nearby agents, and efficiently delegates the heavy-lifting of replication away from Datanet's Central component. Central provides the system’s High Availability. Central itself is a highly available mesh of highly available clusters. This all adds up to real-time distributed p2p replication with strong guarantees and best in class robustness.
CRDT algorithms enable Agents to perform writes even while they are temporarily offline. This is a wonderful feature for systems that want to maximize availability during temporal network failures (e.g. an AppServer with a saturated backend network still processing requests from its frontend network). When the Agent regains connectivity a bi-directional sync w/ Central insures all writes done on the Agent are synced throughout Datanet and all writes done in Datanet are synced to the Agent and the system seamlessly returns to steady state.
A DataCenter is considered down when all nodes in its CentralCluster have died or when the DataCenter itself loses connectivity. Agents that lose connection to their CentralCluster seamlessly cut-over to another CentralCluster, perform a bi-directional sync, and seamlessly return to steady state. This technique also works for multiple simultanerous CentralCluster failures, meaning as long as a single CentralCluster survives the system is fine. Datanet's breakthroughs in robustness are all made possible by CRDT's loose replication requirements.
Datanet's architecture is comprised of a highly available mesh of highly available clusters. Both the HA mesh & the HA clusters support seamless addition of clusters/nodes and tolerate cluster/node failures with no down-time, no data loss, and no loss of write availability. Scaling the HA mesh adds data-centers and replica backup copies to your system. Scaling a HA cluster adds processing & storage capacity to the system.