Storage Processor Claims to Optimize SSD Workloads

Article By : Gary Hilson

Pliops addresses storage stack inefficiencies with its key-value (KV) based storage hardware accelerator that offloads and accelerates data-intensive workloads.

Exponential data growth isn’t new; what’s driving it is, and after 55 years of Moore’s Law, we’re no longer seeing the processor performance increases necessary to keep up. One solution may be a processor geared specifically to data storage.

It’s not just that demand for data storage continues to grow significantly thanks to big data, artificial intelligence (AI), the Internet of things (IoT), and 5G, said Pliops president Steve Fingerhut, but also that so much more data is moving to SSDs with data center SSD usage forecasted to grow a hundred times in the next decade. “SSDs have a lot of great qualities but the primary one is performance over hard drives.” Over the past 10 years, the cost of SSDs has dropped while their performance has improved and capabilities have expanded thanks in large part to the rise of NVMe, he said.

What’s not keeping up, however, is processor performance, and the result is a bottleneck, said Fingerhut, as companies such as Intel and AMD keep adding cores which add cost and power with diminishing returns. “As you add more and more cores, you’re sharing the same memory bus. That’s a big problem.” All this growing data is getting stored on fast SSDs, he said, “but it’s all being driven through a processor which isn’t really evolving all that quickly.”

Pliops Storage Processor (PSP) is a key-value (KV) based storage hardware accelerator that can work with any SSD to offload workload performance and optimize SSD usage. (Source: Pliops)

Pliops’s solution to these “storage stack inefficiencies” is data acceleration technology geared for specific workloads within databases and software-defined storage, which the company estimates is around half the spend in an enterprise and data center infrastructure. The Pliops Storage Processor (PSP) is a key-value (KV) based storage hardware accelerator that enables cloud and enterprise customers to offload and accelerate data-intensive workloads so data centers can continue to scale while reducing computational load and power.

The PSP not only improves workload performance but also optimizes SSD usage, said Fingerhut. Typical use cases include mySQL databases, favored by Facebook, as well as the increasingly popular, yet expensive DRAM-based Redis. Databases normally rely on a storage engine that “amplifies” the data through the sorting and indexing because it will write to the SSD 40 times more than the actual data requested by the application, he said. “You end up using more space on disk, which consumes more of your SSD, and that can be in the two to seven times range. All those amplification effects are due to the fact that the architects are trying to limit how much of the CPU it consumes.”

Fingerhut said how the PSP addresses storage and database-specific processing challenges is much like how workload-specific accelerators have been developed for AI and machine learning applications, which have the benefit of being a relatively new workload. In general, the technical benefits of the PSP for databases include anywhere from three times to ten times as many queries per second and a 93% reduction in query latency, while also reducing power and SSD capacity requirements, he said, which translates into reduced costs.

Pliops isn’t selling flash nor is it selling SSDs. Fingerhut said the PSP can work with any SSD and any interface, and the company is also looking at implementing its technology in FPGAs. He noted both Amazon and Microsoft Azure have recently announced their own database accelerators. “It’s a little bit different focus than what we’re doing but it shows that the race has started. We’re all moving in that same direction.”

Tom Coughlin, president of Coughlin and Associates, said the trend toward offloading functions to specialized computational storage accelerator technologies from other processors such as the CPU reflects the reality that faster processors aren’t being built as quickly as they used to be. “Specialized processors have actually become a big deal.” Other examples include GPUs, tensor flow devices, and other specialized processors for networking, he said.

Released last year, Samsung’s KV Stack is a combination of key-value (KV) SSD and the corresponding host software such as device drivers and associated libraries. (Source: Samsung)

In the case of Pliops, the company’s PSP uses a specialized implementation of KV technology that has actually been around for a while, noted Coughlin, and was even used with a line of Seagate hard drives five years ago that never took off. More recently, companies such as Samsung have released KV SSDs with special flash translation layer software, and the Storage Networking Industry Association (SNIA) has released an open standard for key-value application programming. “Key-value is a common approach for object storage.”

Coughlin said the Pliops PSP has come along at a time when many companies are replacing their hard drives with flash, and it will reduce the need for over-provisioning as well as reading and writing amplification. “You get more endurance and more effective capacity on the drive.”

For now, Pliops is a leader in this area in terms of actual products, he said, but an SSD maker could potentially create a competing product or acquire the company to add its capabilities as a feature in a larger product offering.

Leave a comment