A couple of years ago, the situation seemed clear: solid state storage would operate as a very small top tier, with hard disk drives (HDDs) holding the bulk of data. But that may be changing as the price of flash continues to plummet.
“Over the first 9 months of 2015, flash component pricing dropped more than 60 percent,” said Michael King, director of marketing, DDN.
This makes it more affordable for vendors to come up with systems that can support more solid state storage. But what is the right balance?
Chadd Kenney, CTO of Americas for Pure Storage, says that solid state drives (SSDs) should be deployed anywhere and everywhere for performance-centric workloads.
“SSDs drive massive business value, enabling better user experience, faster access to information and real-time business analytics,” said Kenney.
To his mind, database workloads tend to be a no-brainer for SSDs due to the highly random, small block transactional nature of their workloads, which can challenge traditional disk-centric technologies. Virtualization is another use case, he said, that is optimized by flash as many applications and workloads are mixed on the same data store, making IO activities random. Virtual Desktop Infrastructure (VDI), too, benefits from abundant flash in order to guarantee a smooth user experience.
That said, Kenney is realistic. He noted that without data reduction, solid state storage carries a premium over HDDs. Therefore, it is wisest to apply SSDs in volume in situations where the dataset is most reducible. With deduplication and compression implemented, the cost of SSDs in this case drops considerably, as far less space is consumed.
“Some are surprised that even tier 2 workloads can become cheaper than disk,” said Kenney.
Eric Herzog, vice president, product marketing and management, IBM Storage Systems and Software Defined Infrastructure, is another firm advocate of solid state storage. He said that NAND flash-based storage, including SSDs, provides benefits in nearly all enterprise application environmentsTo deploy flash, he said, is to capture and retain competitive advantage.
“Enterprises today don’t simply want to store data; they want to derive more value from their oceans of data,” said Herzog. “They want to detect and prevent fraud in real-time, gain insight into customer trends, increase productivity, lower operational costs and capture greater market share. They are adding flash storage to help accomplish business objectives.”
He made the case that the purchase cost (CapEx) of flash storage has fallen until it is now close to the cost of HDD. And flash offers operational cost (OpEx) advantages: it consumes less power, requires less HVAC, takes up less floor space, and is more reliable that HDD, according to Enterprise Strategy Group’s Economic Value Analysis.
That’s why IBM now recommends deploying flash for all active datasets. HDD and tape storage can be used for less active or cold data. The more active a dataset, the more it benefits from flash, said Herzog. Inactive datasets, on the other hand, may see cost advantages from being stored on other media.
But results are somewhat dependent on the underlying infrastructure. As the majority of conventional enterprise storage systems in service today were designed to optimize the performance characteristics of HDDs, those deploying flash storage may not initially see the benefits they expect. You have to align the systems in operation to the strengths of solid state storage.
“Everything from operating systems to RAID configurations may introduce hidden bottlenecks or added system complexity that kills the performance and dilutes the latency benefits offered by flash storage,” said Herzog. “Adding a few SSDs to a legacy disk enclosure will likely not accelerate business-critical applications as much as desired.”
Solid State Storage vs. Hard Drives: In-Memory
A major storage boost can be obtained by employing flash with in-memory applications, said Kevin Wagner, vice president of marketing, Diablo Technologies. The primary applications where storage or memory are bottlenecks, he said, include caching, big data analytics and databases.
“Flash should be implemented to accelerate enterprise applications and should be located as close to the processor and system memory as possible where, ultimately, all application data must reside for the application to act upon,” said Wagner.
Flash, then, is blurring the line between caching and storage. It is increasingly being placed outside of the storage array and closer to compute to further reduce latency and increase application performance by orders of magnitude.
“This move will further blur the line between caching and storage, as the flash layer will naturally become more persistent as performance-intensive applications rely upon this ultra-fast tier to deliver results at speeds that outpace their competition,” said King.
Solid State Storage vs. Hard Drives: Hybrid Best?
With solid state storage being faster than spinning disk, it is inevitable that it finds its way into more and more systems, particularly with pricing falling so much. While it may be ideal for virtualization and applications where throughput and latency are paramount, it isn’t a panacea for performance. It can even be the case that an expensive all-flash array with a traditional LUN or volume-based architecture can be slower than a VM-aware hybrid (flash plus HDD) array that is designed for modern workloads, said Chuck Dubuque, senior director of product marketing at Tintri.
“Most enterprises need to balance cost and performance, and right now a hybrid array is a better choice for most workloads,” said Dubuque. “Look for hybrid-flash arrays with high flash hit rates (greater than 90 percent), otherwise you are not getting a good balance of flash performance and HDD capacity.”
The hybrid use case is backed by Mark Lewis, chairman and CEO of Formation Data Systems. He said that SSDs are excellent for high I/O applications and workloads where very low latencies are demanded, especially for high transaction database and online transaction processing (OLTP) applications such as high-volume financial trading systems where every transaction is critical and latency tolerance is measured in microseconds. This doesn’t mean, however, that hybrid is the only use case. He sees some definite instances when the SSDs work best in a hybrid configuration with HDDs. But in other instances workloads may dictate that the storage nodes are pure flash as found in all-flash arrays.
“For workloads that occasionally burst and require additional I/O processing power, hybrid HDD solid state storage configurations can be preferred if the technology has the software logic to intelligently tier the HDD/SSD storage and dynamically place the workloads to the appropriate tier based on pre-defined policies for optimizing performance and/or cost,” said Lewis. “HDD is used when workloads need to be optimized for capacity and cost.”