Database High Availability Simplified

FlashGrid Architecture Highlights

  • Primary shared storage based on standard NVMe PCIe SSDs
  • Physical storage located inside the database nodes (converged nodes) or in separate storage nodes
  • Standard x86 servers or VMs used as database and storage nodes
  • FlashGrid software manages SSD devices and connectivity, integrates with Oracle ASM
  • Oracle ASM manages data, volumes, mirroring, snapshots
  • 2-way or 3-way mirroring of data across separate nodes
  • Fully distributed architecture with no single point of failure
  • Choice of 10/40/100 GbE or InfiniBand/RDMA for network connectivity
  • FlashGrid Read-Local Technology minimizes network overhead by serving reads from local SSDs at the speed of PCIe

Shared Access

With the help of FlashGrid software each ASM instance can access each of the SSDs in the cluster.



Converged Nodes or Separate Storage Nodes

In most environments a converged configuration with storage located inside the database nodes is optimal. The following picture shows an example of such configuration with four database nodes.

A configuration with storage in separate dedicated storage servers may be preferred when database nodes do not have enough room for SSDs, for example, with blades or 1U database servers.

High Availability and Data Mirroring

FlashGrid has a fully distributed architecture with no single point of failure. FlashGrid leverages Oracle ASM’s existing capabilities for mirroring data. In Normal Redundancy mode each block of data has two mirrored copies. In High Redundancy mode each block of data has three mirrored copies. Each ASM disk group is divided into failure groups – one failure group per node. Each disk is configured to be a part of a failure group that corresponds to the node where the disk is physically located. ASM makes sure that mirrored copies of a block are placed on different failure groups. In Normal Redundancy mode the cluster can withstand loss of one (converged or storage) node without interruption of service. In High Redundancy mode the cluster can withstand loss of two (converged or storage) nodes without interruption of service.



FlashGrid Read-Local™ Technology

In converged clusters the read traffic can be served from local SSDs at the speed of the PCIe bus instead of travelling over the network. In 2-node clusters with 2-way mirroring or 3-node clusters with 3-way mirroring 100% of the read traffic is served locally because each node has a full copy of all data. Because of the reduced network traffic the write operations are faster too. As a result, even 10 GbE network fabric can be sufficient for achieving outstanding performance in such clusters for both data warehouse and OLTP workloads. For example, a 3-node cluster with four NVMe SSDs per node can provide 30 GB/s of read bandwidth, even on a 10 GbE network.