Hybrid flash/disk storage systems, in which data is written first to a flash cache in the server and then later peculated to a traditional storage array, can provide significant direct savings in environments supporting more than 700 IOPs per Tbyte. And hybrid arrays can benefit SMBs as well as large enterprises. Those are the primary conclusions of an in-depth Wikibon research project into performance differences between true hybrid storage and traditional arrays with read-only cache, presented in David Floyer’s latest report, “Hybrid Storage Poised to Disrupt Traditional Arrays.”
The savings comes partly from the order-of-magnitude read-write speeds of flash-first hybrid storage, in which data is written directly to the flash cache, as contrasted by the traditional array architecture in which data is first written to disk and then copied to the flash cache. This advantage is magnified in virtualized environments where control of reads and writes rests in the virtualization layer, and traffic from multiple applications is combined. This traffic appears to the storage layer as random. Flash is optimized for such random traffic while disk by nature is much more efficient handling long consecutive reads and writes.
The result of this huge performance advantage is that the number of cores in a typical high-performance database can be reduced while still meeting QoS requirements for the database. For instance, Floyer writes, a planned eight core traditional database infastructure can be reduced to five cores by using a hybrid system. Given that licenses for SQL Server and Oracle databases cost $15,000 to $20,000 per processor core, this can save $45,000 to $60,000 in licensing.
A VM-aware hybrid storage architecture also greatly simplifies database management because VM storage objects are mapped directly to objects held in the storage array. The software-layer management can provide detailed response-time performance measurements to the DBA. The study found that this reduced the time required to manage these systems to about five hours per week, compared to 10-to-20 hours per week for traditional systems running similar loads.
Databases under 700 IOPS per Tbyte, and lower performance systems in general, however, probably will not benefit sufficiently to justify the higher purchase cost of flash systems. They are better off, Floyer writes, on lower-cost all disk systems.
Adding read-only flash cache to traditional storage arrays will increase performance. “However, Wikibon does not believe that this implementation will provide the same cost and performance advantages of a full implementation of a flash-first VM-Aware hybrid architecture.”
Floyer concludes that over time flash costs will continue to decrease, and flash-first arrays, either hybrid or pure flash, will become the standard for high-performance storage. As a result, he recommends strongly that CIOs, including those of SMBs and corporate divisions, specify low-latency flash or hybrid arrays in the RFPs for all high-performance storage systems where the anticipated performance characteristics exceed the minimum 700 IOPS per Tbyte.
As with all Wikibon research, this report is available in its entirety on the public Wikibon Web site. IT professionals are invited to register for membership in the Wikibon community. This allows them to comment on research and publish their own Professional Alerts, tips, questions, and white papers. It also subscribes them to invitations to the periodic Peer Incite meetings, at which their peers discuss the solutions they have found to real-world problems, and to the Peer Incite Newsletter, in which Wikibon and outside experts analyze aspects of the subjects discussed in those meetings.