Server-Side Flash: Under New Management, But Is It Ready for Prime Time?
For those who support it, server-side flash seems like the answer to complex networking problems in the enterprise. After all, what’s not to like about literally terabits of storage at or near the processing center with nothing but a high-speed PCIe bus between them?
Not so fast, say the traditional SAN and NAS lobby. Giving up on traditional storage architecture also means giving up on a number of critical features like RAID protection, dynamic resource allocation and automated failover and recovery.
But that argument is coming under increased pressure now that new flash management technologies are hitting the channel.
A case in point is Sanbolic’s new Melio5 software. The package brings a number of SAN-like features to flash-based storage architectures, including multi-node scalability and availability, remote replication and snapshot capability. At the same time, the system delivers upwards of 650,000 IOPS per server at a fraction of the cost of a traditional storage system. The goal, company execs say, is to convince storage professionals that server-side flash, solid-state, and even hard disk storage is ready to perform primary storage duties.
At the same time, Virident is out with new connectivity software aimed at improving the pooling capabilities and availability of server-side flash. The FlashMAX Connect suite is intended to supplement the company’s FlashMAX II platform to create a high-performance storage tier for database and virtualization applications. The system includes modules for shared storage and data management that seek to foster cooperation between flash management and wider storage management systems. This allows functions like data sharing across remote servers and flash scalability independent of compute scalability, so that a single PCIe card can operate across multiple servers.
Enterprise flash management is also drawing newcomers to the field. Start-up PernixData recently came out of stealth with the Flash Virtualization Platform (FVP), which the company pitches as a solution to the increasing storage requirements of virtualization. The idea is to provide a software-defined, scale-out storage environment that allows the enterprise to seamlessly leverage flash as a clustered resource, without forcing changes to VM architectures or existing storage platforms. In this way, the company hopes datacenter managers will be able to leverage high-speed flash resources to support VDI and other applications without having to dramatically increase traditional storage infrastructure.
Even firms perceived to be at the forefront of the enterprise flash storage movement are moving quickly to shore up their management capabilities. Fusion-io recently bought out British storage management firm ID7, developer of the command processing stack SCST. On the surface, the deal puts Fusion-io in control of a key enabling technology for open-source storage solutions. However, it doesn’t take a lot of imagination to picture how it can be used to speed up Fusion’s flash-based storage technology as data is passed to and from wider data infrastructure.
So are we ready to write off traditional networked storage? Not quite. True, SAN/NAS architectures are costly and complex, but they have a proven track record when it comes to enterprise-class data environments. Flash is still widely seen as a high-speed cache solution that enables existing data infrastructure to keep pace with increasingly complex virtual and mobile endeavors
The thing to watch is whether new datacenter construction will continue to rely on SAN/NAS or pursue more modular approaches that stress integrated compute/storage/networking components. If new flash management capabilities bring SAN-like features to server-side solutions, they may not cause the enterprise to swap out storage networks en masse, but they could be the solution-of-choice for future infrastructure deployments.