The first wave of all-flash arrays was typically purchased to address a specific performance problem like a poor performing database application or to allow a VDI project to live up to user expectations. Most all-flash arrays successfully addressed, and in many cases obliterated, these performance problems. In fact, they often have performance to spare but as IT professionals began to add more workloads to these arrays, weaknesses in initial flash designs are exposed. NexGen recently announced an all-flash array that addresses these challenges and promises to allow data centers to do more with their all-flash investment.
Designing The Multi-Purpose All-Flash Array
While not identical, the NAND between all-flash arrays is very similar. It is the package that surrounds the all-flash array that differentiates between the various systems on the market today. The all-flash array is a combination of hardware and software, and how they work together as a system determines the level of workload variability.
To extend an all-flash array beyond its original use case requires that the all-flash array is cost effective so that a high number of the workload’s data can be stored on it. Many flash vendors will deliver cost effectiveness by leveraging data efficiency techniques like data deduplication and compression. Using these techniques is fine as long as there is redundant data available. Also important is leveraging the most affordable flash tier possible. This means making the solution cost effective from a raw hardware perspective, you can’t count solely on the “fuzzy math” of deduplication and compression to make the solution competitive.
In addition to being cost effective the all-flash system needs to provide predictable performance. Predictable performance has two components. The first is consistent performance for a wide variety of workloads. The all-flash storage system’s performance becomes unpredictable under load when less expensive consumer flash is used instead of enterprise flash. This unpredictability is especially evident as the type of workload increases and becomes more variable. The problem with using enterprise class flash exclusively is that the solution becomes too expensive to support all but the most demanding workloads. The mixed workload problem can be solved if all-flash arrays start to leverage several memory tiers. The goal would be to use flash – a small amount of enterprise flash or even DRAM to store the data initially and then move the data to consumer grade SSDs as it moves to a read-only state.
The second component in predictable performance is making sure that it will provide the exact performance to a given workload no matter the situation; in other words, quality of service (QoS). Is QoS in an all-flash array even necessary? After all, the systems are so fast that they can often generate more raw IOPS than most data centers need. Again, the Flash NAND often is not the bottleneck. Instead, the “package” or system that surrounds the NAND chokes the system. An all-flash array will often be bottlenecked by lack of storage CPU or network I/O resources instead of media performance. QoS allows the allocation of all-flash array system granularly so that the highest performance is reserved for the applications that need it.
Introducing the NexGen Multi-Tier All-Flash Array
NexGen has been a provider of hybrid systems, flash, and hard disk, for several years. Their hybrid systems are unique in that the flash tier is PCIe based allowing them to deliver very high performance at very low latencies. Their hybrid system also has QoS that allows the IT planner to set specific IOPS requirements for each application.
With the announcement of the N5-1500 and N5-3000, NexGen is replacing the hard disk tier with consumer-grade flash drives. The PCIe tier acts as a shock absorber to the consumer grade flash tier, lowering the number of writes the tier has to handle. In addition to a secondary flash tier, NexGen has also added a RAM based tier for read acceleration. While the hardware promises to deliver 300,000 IOPS of raw performance, the QoS software promises to apply that raw performance judiciously so a high number of workloads can be centralized on the array. The result should be an all-flash array that can support the breadth of the applications in the data center without fear of unpredictable performance and do it with utmost efficiency.
The N5-1500 has 2.6 TBs of PCIe flash capacity and 15TBs of raw SSD flash capacity in its starter package. It is expandable to 60TB of SSD flash capacity. The N5-3000 also has 2.6 TBs of PCIe flash but starts at 30TBs of capacity, also expanding to 60TBs. Both systems have 96GB of RAM, of which a portion is used for read acceleration.
One of the motivators for choosing all-flash arrays over hybrid arrays is predictable performance. The problem is that the addition of new workloads creates performance variances that lead to predictability problems under load. Part of the cause of this problem is that most flash arrays use a single tier of flash, another is that they have no or limited quality of service (QoS) capabilities to make sure that important apps get the performance they demand. NexGen, with the N5-1500 and N5-3000, has addressed both of these challenges and promises a single system that can deliver consistent performance to all the applications in the data center.