Server-side Flash is quickly emerging as the go-to solution for enterprises looking to get more performance out of their virtual infrastructure. But while it excels in throughput, it lacks some of the key features needed in a storage environment: mainly, the ability to aggregate disparate components into a shared resource pool.
Lately, however, a number of intriguing developments have come out of the lab with promises to turn Flash modules into highly scalable cache solutions or even full-fledged virtual storage environments in their own right.
VMware, for instance, has incorporated its vCache system into vSphere 5.5 under the new name Flash Read Cache (FRC). The system allows individual hypervisors to create their own logical storage environments using available SSD or PCIe-based modules. The idea is to provide a scalable, high-speed cache environment so the hypervisor doesn’t over-consume either network or storage resources on the SAN, hampering performance and forcing enterprises to over-provision existing physical infrastructure. On its own, FRC is limited to the Flash storage available on a single physical host, although it is possible to transfer both the hypervisor and its cache to another Flash-enabled host using vMotion.
Wouldn’t it be better, though, to have a fully virtualized server-side Flash environment in which resources could be pulled from host to host at will? That’s the intent behind PernixData's new FVP flash hypervisor software, optimized for NAND environments. The system works by clustering multiple hosts and then establishing a Flash hypervisor that can pick and choose available storage from any module in the cluster. There is no limit to the number of virtual appliances in the Enterprise edition, although the SMB version supports only four, and it provides full read/write acceleration for all applications. FVP is optimized for VMware environments, with built-in support for vCloud Director, vMotion, Horizon View and Site Recovery Manager.
Meanwhile, Virident Systems is on a similar track with its FlashMAX Connect suite that works with the company’s FlashMAX II server-side platform. The system can be used to build Virtual Flash Storage Networks with either local PCIe cache or through block-level access across servers. In this way, managers are able to scale their Flash capacity independent of server resources, potentially delivering from 500 GB to 24 TB for each server. The system supports vHA, vShare and vCache software modules, which can be run separately or as an integrated Flash management platform.
The irony in all this, though, is that even as most of the Flash storage industry is working on optimizing SSD and PCIe deployments, others are already looking at even higher-speed solutions. Canada’s Diablo Technologies, for one, recently unveiled an entirely new system dubbed Memory Channel Storage (MCS) that puts Flash directly in the DIMM socket. This boosts performance by several orders of magnitude by circumventing the storage subsystem entirely and leveraging the memory controller’s parallel processing capability. As well, you now have the ability to scale system memory into the terabyte level, which provides ample support for all but the largest of workloads even if the technology does not yet support multi-host resource aggregation.
Such considerations are probably best left to the future, however, as DIMM-based Flash itself is barely off the drawing board. In the here and now, enterprises are concerned that server-side Flash solutions will quickly generate the same kind of data silos that virtualization was supposed to remove. With new Flash pooling technologies hitting the channel, it seems that the goal of fully federated, high-speed data infrastructure is quickly becoming a reality.