All-Flash Storage and Flash-Сentric Hybrid Storage
By Sergey Platonov
What is All-Flash Storage and Flash-Сentric Hybrid Storage?
How exactly do data storage systems support Flash? Can a general-purpose data storage system be converted into All-Flash Storage?
Readers who closely follow marketing materials from data storage tycoons will probably answer that All-Flash Storage must include deduplication and in-flight compression to decrease storage device wearing. Somebody will mention the need to balance writing operations by the size of flash-memory erase-block. Select best practices advise on avoiding the read-modify-write pattern in favor of redirect-on-write. Others will recall log-structured file systems and their interaction with Flash translation layer, still others will bring up new coding schemes. Besides, everyone will emphasize the importance of TRIM/SCSI UNMAP support.
In reality, development of a Flash-based data storage system always needs an all-round review of the storage architecture. Technologies are secondary!
A revolutionary decrease in NAND Flash costs is expected around 2017–2018. Standing out from the crowd, Seagate suggests that NAND Flash will still maintain a way higher price tag compared to HDD storage for a few years to come. Since this company offers both types of storage devices, they must know what they are talking about. Seagate has under their belt the right tools to squeeze ten-fold disk capacity growth out of the old good HDD.
All-Flash and Flash-Сentric Hybrid Storage will sure recruit fans among those who shop for sustainable high performance and do not get stick-shocked easily. Like classic storage, these arrays must commit to a high-rated QoS. System response to 99,9% of requests must not exceed a specified showing, whatever the circumstances. According to Wikibon, the upper limit will drop down to 500 microseconds in the next 3 years. Now it’s around 1–2 milliseconds.
As Henry Baltazar from Forester Research infers, if your vendor doesn’t have Storage QoS on the roadmap, it’s high time to quit on that vendor. Baltazar highlights guaranteed performance level (IOps and MBps) for specific LUNs and virtual machines.
The RAIDIX team views this issue from a slightly different angle. Response time is great, however, QoS is not about limiting speeds for “unimportant” users. It’s about keeping up sustainable productivity levels for all user groups – and balancing performance across versatile workloads.
On developing our RAID stack, we focused on data recovery in a matter of nanoseconds. To ensure QoS on reading, we use Advanced Reconstruction, which allows to re-calculate data faster than it could be read from physical disks.
An unprecedented feature, RAIDIX’s QoSmic allows to align performance with various workload patterns and prioritize applications based on their behavior patterns – with the use of AI principles.
At the bottom line, our definition of All-Flash Storage is guaranteed performance leveraging millions of IOps with microsecond latencies. At any given moment and whatever happens to the system.