2020 Storage Trends from RAIDIX

«Revolutionary changes are unlikely, but uncommon technologies will gradually become mass-scale», — Sergey Platonov, VP Strategy, RAIDIX, shares his tech forecasts for the year.

At RAIDIX, we develop software for storage systems used worldwide in M&E, HPC and Video Surveillance projects. In 2019 we delivered (among everything else!) new releases of RAIDIX 4.X for HDDs-based storage and RAIDIX ERA 3.0 for NVMe drives. We have decided to look into the future and try to foresee what will happen next.

Hard Disk drives (HDD): the total volume of released HDDs will grow. The demand for SMR-compatible software will increase.

Sales of HDD, expressed in a number of drives, will drop, but the total storage amount is going to increase nonetheless. With less performance density (IOps/TB) there are not so many HDD application scenarios left.

We are going to witness the Renaissance of Shingled Magnetic Recording technology (SMR), and the majority of HDD-focused software will start supporting SMR soon. Thus, Western Digital has already opened up 20 TB drives and is planning that at least half of drives to be released within the next 3 years will be SMR-based. We are planning to adapt RAIDIX to this technology as well.

Solid State drives (Flash): awaiting for new hybrid solutions consisting of rapid memory (3DXpoint, for example) and large QLC-based drives (including ZNS support).

In 2019, we met QuantX from Micron, an analog of Intel Optane, and Toshiba introduces its XL-flash. New servers and products will support PCIe 4.0, doubling the bandwidth. On the other hand, new form factors and the higher number of bits per cell and layers of 3D NAND memory will allow compact placement of large volumes of Flash, set for read operations. RAIDIX team will continue further development of RAIDIX ERA, which perfectly works together with these devices and shows great results during both read and write operations.

Zoned Namespaces technology (ZNS) will further increase volume of NAND drives. In 2020, we expect to emerge of the first ZNS drives while the market badly needs software that works with ZNS, brings out all performance of write-focused and low latency devices (Optane, QuantX, XL-Flash) and creates an effective virtual pool that works on top of these two levels.

New devices: expecting new ready-to-use devices of the computational storage class.

Computational storage drives allow moving some tasks as close to the data as possible and unload CPU host. SNIA has already been working on standardizing the approach, and devices that speed up applications (such as Hadoop and RocksDB) have been already introduced. SmarNICs will partly seize the computation tasks from the servers, and we are going to see a major increase in demand for these devices when building and upgrading data centers. Separate RAIDIX RND team is working on the compatibility of RAIDIX ERA and SmartNICs.

We think Mellanox and Broadcom will step into the bright spotlight, with more compatible software for their devices from independent developers (and by the way, we have already met a number of new products from these vendors).

Non-volatile memory: storage and SDS vendors should really take care of the deep integration of their products with PMEM devices.

With the advent of Intel Optane Memory, a growing number of software companies announce, that this type of memory can be used both to reduce level of delays when accessing permanent storage, and to increase RAM. The main question for 2020 is this: will someone be able to compete with Intel and release a mass and inexpensive product employing an alternative technology?

SAN: in 2020, most companies producing All-Flash will start supporting NVMe over Fabrics access interfaces. RAIDIX will introduce NVMe over Fabrics compatibility in one of the next releases.

Mainly these storage systems will be demanded by companies in these markets and with these tasks: Media (4K and 8K video), AI (that is becoming a general-purpose technology), large databases, Real-Time Analytics, containers and Kubernetes, HPC. We expect that cluster FS and parallel FS will provide full NVMf, as well as the corresponding level of performance.

Architectures: Composable Storage will continue to evolve.

Recall that Composable/Disaggregated Storage involves allocating separate pools of different resources, united by two types of networks, and combining them into virtual devices to solve certain tasks. We are likely to see new hardware models and new products built on the basis of this approach very soon, and new data centers will gradually use these products in their infrastructures.