Hybrid storage systems create a performance concern for IT professionals. The system may respond too slowly to user or application requests for data. If the hybrid system places data on the wrong storage tier at the wrong time, it validates the concern. An all-flash array makes all IO fast and reduces the chance of user complaints. An all-flash array will also significantly reduce the available IT budget.
For most organizations, a hybrid system makes more sense than an all-flash array, if the design overcomes the performance concern. As we discussed in blog 1 understanding the math behind the organization’s data enables the organization to create a hybrid storage system that performs the same as an all-flash array but maintains all the price advantages of hybrid.
The first hybrid arrays were a combination of a small high performance SSD flash tier (3%) and a large medium performance hard disk tier. These systems were successful and in the next generation of hybrid systems vendors pushed flash to its limit. The second generation of hybrid systems used hard disk drives with a slightly larger flash tier (5%) and much higher capacity (though slower).
Second generation hybrid systems achieve an impressive price point while maintaining some flash performance. However, when placing these hybrid systems into production, it became obvious there was too much disk capacity and too little flash. The lack of available flash forced too many IO requests to come from disk. The larger disk drives, meant fewer drives and slower per drive performance. Not only did these systems count more on the disk tier, the performance of that tier declined. This increased use of a slower hard disk tier for IO led to user disappointment, which in turn led IT to sacrifice their budget and purchase flash-only systems.
The Hybrid Solution
Instead of abandoning the hybrid model, and sacrificing their budget, IT needs to rethink the hybrid design. IT has two options. First, they can increase the amount of flash used in the system. IT planners can cost effectively create a much larger flash tier, making flash 25% or more of total capacity, because the cost of flash storage continues to decline. Using the deep math we discussed in blog 1 it is safe to assume that a 25% flash tier would almost always serve data from flash storage, making reads from the disk tier extremely rare.
All-Flash Hybrid Arrays represent the second option. Organizations with high data turnover or organizations that remain concerned about user dissatisfaction can look for systems that use a combination of high performance flash and high capacity flash.
Typically, the high performance tier is NVMe Flash and the high capacity tier is SAS based flash with very high capacities per drive. The SAS tier is disadvantaged by its network interconnect, SAS, but the NVMe tier will handle most of the IO load. The lower performing SAS interconnect can easily handle the reduced IO load, meaning that reads from it perform almost identically to the busier NVMe tier.
Hybrid storage systems still make sense. Organizations need to look for systems with the appropriate amount of flash capacity relative to disk capacity or select an all-flash hybrid design. In either case, the customer benefits from a lower cost alternative to a modern NVMe based all-flash array without giving up performance.
In our next blog we explain, again using deeper math, how an object storage system supporting a hybrid flash array can meet the response times that the organization needs while also positioning it to better protect and maintain data compliance.