EMC is actively encouraging data center managers to take an aggressive, intelligent approach to extending the benefits of flash through the promotion of Flash PowerPacks. These value-priced drive packs provide a significant opportunity for customers to save on their flash storage purchase, but also provide an incentive to expand flash use in their hybrid VMAX and VNX2
The initial deployment of solid state drives in general purpose arrays was tempered by the steep cost per GB, combined with the challenge of ensuring the most performance-critical data received the benefits. Users could take some comfort in the mantra “a little flash goes a long way.” With generally declining prices for NAND storage, combined with more economical eMLC technology and auto-tiering software like FAST VP for FAST VP tiered storage, users can now explore the benefits of increasing the amount of flash in their VNX and VMAX systems beyond the 5% (or less) of capacity that many use today.
With PowerPacks users are encouraged to explore the benefits of moving to a higher percentage of flash storage. These benefits include: greater average IO performance per drive (10,000 IOs per second for an SSD as compared the approximately 120 IOPS for a typical 10,000 rpm HDD); the ability to support data types that remain “hot” for longer periods; and the availability of flash capacity for dedicated use with demanding applications such as OLTP and databases.
Flash storage is well on its way to being the de facto answer for storage needs where performance is concerned, however cost remains a primary concern for data centers that must support a broad array of users, applications and data types. As long as spinning drives offer significantly better cost per GB of data stored, users with diverse application environments and large amounts of data will continue to depend on hybrid SSD/HDD storage, and endeavor to understand how much flash storage will provide them the optimal amount of performance density, while retaining the economic benefit of a majority of spinning disk.
For environments where flash as a tier is accelerating a mix of workloads, it may not be obvious how much flash to deploy for best results, where the greatest benefit occurs, or whether there is a point of diminishing returns. In fact, with a variety of all-flash array products available, users may simply assume that “more is more.” By making a couple of simplifying assumptions, we can demonstrate why the ability to cost-effectively increase the amount of flash as a percentage of total storage can have a significant impact on the average performance of a hybrid storage array. This same analysis also shows that, for environments where new to old (data) also means hot to cold (access frequency), close to all-flash performance can be had for considerably less than an all-flash price.
Our simplifying assumptions:
- Data is growing by some amount annually – we’ll use 60%: this number has become akin to the Moore’s Law of storage;
- That new data is the most active, and activity decreases as the data ages;
- The cooling off of data is similar to radioactive decay; and
- Flash coverage required is determined by amount of new data and how long it stays active.
Another useful chart would be the one that shows the impact of data aging on the minimum amount of flash coverage needed. In Figure 2, it appears that the relationship is linear – data that cools more slowly will require more flash to maintain performance. For our example, a minimum of 9.25% of total capacity should be SSDs in order to cover the 90-day period, after which 90% of data is inactive.
The question remains: what is the relative benefit of adding more flash? And further, how can users leverage PowerPacks to tap into those benefits? One answer to that is to look at how average drive performance (across the hybrid mix of SSDs and HDDs) improves as the percentage of flash in the system is increased. After all, if SSDs are almost 100x better at handling random IO, then we want our average drive to perform much closer to an SSD.
The answer is actual two rather surprising answers: 1) the benefit to average drive performance actually accelerates as flash is added; and 2) users can achieve all-flash performance far short of an all-flash configuration. This latter result hinges on having the declining access rates with time to leverage the HDD half of the hybrid equation.
Look at the curve in figure 3. Up to a flash level of 10% of total capacity, there is definitely a benefit, since average drive performance is up 10X from the base of a 120 IOPS for 100% 10,000 rpm HDDs. However we are still about 90% below all-flash IO performance of 10,000 IOPS per drive, and we are still on the flatter part of the curve. Push beyond 10% however, and we enter the flash impact zone:
The steepest part of the curve where each percent of flash added yields the biggest increase in average drive IOPS, i.e., the biggest bang for the unit of local currency.
Conclusion #1 – The typical 5% flash deployment is getting results, but is also operating below the point of maximum incremental performance benefit from SSDs. Flash PowerPacks make it cost effective for customers to break into the flash impact zone by offering up to 4% of additional flash at the same price paid for individual drives.
Conclusion #2 – The second point to take away from figure 3 is that, beyond 25% flash, hybrid users are approaching effective all-flash performance, achieving almost 90% of all-flash performance with only 25% of their system capacity in SSDs. This is enabled by the hot-to-cold behavior of their application and user data, combined with the intelligent data placement provided by FAST VP auto-tiering.
Different assumptions for data aging and rate of data growth will alter these curves, however the general behavior shown here is typical for a wide range of inputs. Users looking for greater performance and value from their hybrid VNX or VMAX platform should look at what Flash PowerPacks can do for their environment.