Efficiently and cost effectively deliver the highest possible performance
Because 66% of storage resources are wasted on three-way replicas, all-flash hyperconverged appliances have been cost prohibitive. Gridstore’s next-generation architecture is the first and so far only design that eliminates this problem and makes all-flash HyperConverged Infrastructure practical.
Gridstore uses flash in an end-to-end architecture to both accelerate performance and to eliminate as much as 80% of storage traffic, returning CPU cycles to hosts. Gridstore All-Flash HyperConverged Nodes use flash as a local cache and as a high-performance storage resource. By putting a large flash cache (1TB) close to the workload, the entire working set of
approximately 30 server VMs or 200 desktop VMs can be cached locally. Our vController sits in front of this cache, becoming part of a controlled data path with VM granularity. This critical feature ensures the cache resources are utilized according to per-VM policies and not consumed just by the most active VMs. With Gridstore, approximately 80% of I/O (reads) is served from the local low-latency flash cache. This eliminates 80% of storage traffic that would otherwise go across the network and allows storage nodes to focus on serving the remaining 20% (writes). By comparison, an all-flash array must handle 100% of the I/O (reads/writes) in the same controller through the same network connections. Gridstore’s end-to-end distributed all-flash architecture will serve about 80% of its I/O with microsecond latency by not going across the network for reads. When compared to millisecond latency for an all-flash array where all I/O reads/writes incur the network latency, this results in 10X lower latency, which is critical for high performance workloads like SQL, server virtualization, VDI, and big data.