IBM’s recent IBM Flash Systems storage briefing highlighted IBM’s leadership and success in the growing market for computational storage. Computational storage devices process data directly on storage rather than in an external processor, improving performance and throughput, reducing latency and using less power. IBM’s focus is on “computational storage for storage” offloading storage software functions (such as compression) to the storage itself. Their goal is to maintain the smallest footprint, highest density and lowest power in the industry.
Currently in its third generation, IBM FlashCore Modules (FCM) provide the foundation for IBM’s computational storage strategy. IBM FCM are a family of high- performance custom SSD drives residing in block storage using NVMe and high-speed NAND memory. IBM disclosed that about 90% of all SSD’s for NVMe in storage appliances sold to clients are FCM and that they have shipped over 200,000 total units to date. In addition, the company reports that FCM, when used for data compression, offers the highest performance of any data reduction solution in the industry.
According to the Storage Networking Industry Association (SNIA), computational storage is defined as “architectures that provide Computational Storage Functions (CSF) coupled to storage, offloading host processing or reducing data movement. These architectures enable improvements in application performance and/or infrastructure efficiency through the integration of compute resources (outside of the traditional compute & memory architecture) either directly with storage or between the host and the storage. The goal of these architectures is to enable parallel computation and/or to alleviate constraints on existing compute, memory, storage, and I/O.”
Typical functions that benefit from computational storage include compression, database acceleration, content delivery networks, edge computing, artificial intelligence (AI), and encryption. In each of these cases processing occurs directly on the drive, offloading the host processor for improved performance, as well as improving bandwidth by reducing the amount of data going back and forth between the drive and the compute system.
IBM FCM– A Closer Look
If you’re still unsure about the benefits of computational storage, consider this example. The IBM FlashSystem 9500 has up to 100 GB/second read throughput – fast, but can impact servers and networks. 168 GB/second of NVMe bandwidth is inside the 9500 and can be used for functions such as background scrubs. However, there is also 768GB/sec of bandwidth off the NAND flash much of which is unused–providing huge potential for analytics and background processing and saving CPU power for applications that run best on the host.
Third generation IBM FCM uses a standard 2.5″, 15mm dual-port form factor and are available in 4.8 TBu (Terabytes usable) / 22 TBe (Terabytes effective), 9.6 TBu (29 TBe), 19.2 TBu (58 TBe) and 38.4 TBu (116 TBe). The current generation has increased effective capacity using the same compression algorithm but allowing for highercompression ratios. At 100% utilization this translates to a 3 : 1 ratio, while data that is 3.7 : 1 compressible would take 80% physical utilization. This not only increases effective capacity, but in fully allocated mode it also allows more thin provisioning.
Added to the large and extra-large FCM modules is a larger FPGA that increases the number of compressors and decompressors, and provides higher throughput and performance. MRAM (magnetic RAM) has been reduced in order to remain cost competitive and additional savings are realized by consolidating DRAM (dynamic RAM) with an innovative caching technique that allows the use of standard DRAM (lower cost). These features provide persistent storage that isn’t just a repository for data but is also a compute element — with data reduction and computation spread over FCMs.
IBM describes “computational storage for storage” as storage software as an application that runs on x86 servers with failover capability. The first function IBM has tackled is compression. By using the SSD’s in the FCM module for compression, the processor and software are offloaded from doing intensive calculation with the processing being done in-line — with no impact to the stack and data path. Data is written, destaged from cache and compressed, and the read gets decompressed.
What does the future hold? IBM plans to continue its leadership in density and footprint and power reduction. They will build on compression capabilities, and will map additional storage software functions including RAiD, replication, data reduction, monitoring, snapshots — offloading all these functions to the FCM in a distributed fashion.
In addition, IBM plans to use FCM in the future to aid in cyber resilience and ransomware/malware detection. IBM also plans to use FCMs to collect and summarize data of all types to assist in analytics and machine learning and identify trends– comparing new and old data about things such as encryption and looking for data anomalies that indicate a presence of ransomware or malware. Log files can also be scanned to look for various strings. By performing these functions at the lowest level of summarization, bandwidth and data storage is reduced at higher parts of the system.
IBM has differentiated itself in this growing market by providing industry leading density and sustainability, enabling customers to use less hardware and still achieve cost-effective high performance. Updates in the third generation (use of standard DRAM, larger RPGA at high-end) indicate a commitment to continuing to drive cost down while improving performance.
Also, building on its success with offloading compression to the FCM – IBM will do the same with other storage functions. Hence the moniker, “computational storage for storage”. By harnessing the power of the FCM, these functions can be performed with minimal impact on CPU and bandwidth.
With IBM’s history in AIops, that, too, will be an area of focus, particularly in the area of rapidly growing market for ransomware/malware detection.