Accelerating database performance remains a top use case for all-flash arrays, so it makes sense that most all-flash array systems only support block storage. However, only supporting block means that the customer assigns the “raw” volume to a server and then formats that volume with a file system, prior to storing any data on the volume. The customer could also have the database write natively to the raw device. Either deployment method is acceptable for the database use case but the block only all-flash array requires that the customer buys and maintains a separate stack of storage for Network Attached Storage (NAS) and, for customers needing object storage, a separate stack for object storage.
The time for a block only deployment model has passed. All-Flash storage software should provide both file and block services. One of the problems though, with all-flash systems providing both file and block services, is the vendor tends to compromise on one of the two options. Vendors with a file system focus tend to provide block storage as a blob within their file system which impacts raw performance potential. Vendors with a block focus tend to repackage an existing open-source file system solution, which lacks tight integration into the vendor’s solution.
All-flash storage software should provide both block and file services and do so without compromise. Block performance should remain as high as possible, but file services should be completely integrated into the storage software. Integration enables IT to easily provision and monitor both block and file services from a single interface. It is also important that the file system provide high performance as an increasing number of workloads, like high velocity analytics and artificial intelligence, now demand performance similar to block storage.
Time to Re-Think Storage Software
The past five blog have discussed how storage software needs to change to keep up with new capabilities and potential of the hardware. The processors that drive the storage software are now massively multi-core. The storage devices that the software now manages, offer the potential of over a hundred thousand IOPS per drive and very low latencies. Most importantly, the performance demands from users and the demands for cost efficiency from the executive team are both at all-time highs.
The answer is not to continue to buy more and more hardware. Instead IT needs to look for multi-threaded storage software that takes full advantage of multi-core processors and re-evaluates how legacy algorithms like RAID calculations are made to extract maximum performance from new NVMe flash drives. That software, however, also needs to deliver a complete, enterprise class feature set including data protection, automated tiering and flexible hardware selection.
Vendors like StorONE are providing a software-only solution that for the first time can provide block, file and object all in a single unified solution. This enables users to now use their AFA for many more use cases simultaneously while achieving much higher value when purchasing AFA hardware. The solution has repeatedly shown the ability to extract maximum value and performance from the drives used, besting systems from better-known companies.