Since the arrival of HyperConverged Infrastructure, the biggest customer complaint has been inefficient scaling. First-generation products force customers to scale servers at the rate of storage growth. To make matters worse, three-way replication force customers to grow servers at 3X the storage growth rate. For some workloads that are uniform (such as VDI), this is not a problem and many installations have started with these workloads. For many other general purpose workloads, however, storage grows considerably faster than the compute workloads do.
Unlike other HyperConverged Infrastructures, Gridstore is not iSCSI, NFS, or SMB serving I/O.
Rather, Gridstore is a peer-to-peer Grid architecture designed specifically for this purpose, addressing the following critical scenarios:
- Independent Scaling: Independently scale storage only using purpose-built storage nodes that can be deployed incrementally as storage is required. This supports mixing different classes of available storage (flash, hybrid, and capacity) to allow different types of capacity pools to grow to meet different workload requirements or budgets.
- Compute Scaling: Scale compute only using any X86 server that runs Windows. A Gridstore vController can be deployed in less than a minute to any Windows server. That server, in turn, can become a node in the Grid, with full access to all shared storage. Adding SSDs to any of these servers enables the vController to use them to accelerate workloads on servers. Hosts can be physical servers or hypervisors. If they are physical servers, features like end-to-end storage QoS can be applied to deliver precise amounts of IOPS to each of these servers. Hence, the Gridstore architecture allows scaling of both compute and storage separately. Compute can be scaled using any or existing X86 servers while storage can be scaled to fit precise requirements using storage-only nodes.