Storage Switzerland is at the 2019 edition of the Flash Memory Summit (FMS). Each day of the summit, we will be providing a quick summary of our meetings. Day 1 was a busy day for StorageSwiss as we raced from briefing to briefing.
Briefing 1: Pavilion Data
Pavilion Data, whom we covered in our “Designing Storage for MongoDB, Spark, MySQL, and Cassandra” briefing note, is a storage solution designed to optimize fully the performance of Non-Volatile Memory Express (NVMe) Flash for these modern workloads. Pavilion’s architecture looks more like a director class switch than a storage array. It has ten dual controller line cards. Each card has NVMe flash and four 100GE ports. Designed to be highly available, the unit has four redundant power supplies and two management modules. It supports up to 72 NVMe flash drives with a starting capacity of 14TB and is expandable up to 1PB. Each controller has access to any drive.
In addition to their expected success in web-storage / hyper-scalers, Pavilion is also seeing success in markets like financial services, Government, High Performance Computing (HPC) and Media and Entertainment. They are also looking to make a push into the energy sector, which in many ways resembles the HPC market.
Briefing 2: NVM Express
NVM Express is an open collection of standards and information to expose fully the benefits of non-volatile memory in all types of computing environments from mobile to data center. The design of NVMe, from the ground up, is to deliver high bandwidth and low latency storage access for current and future NVMe technologies. NVMe as a drive technology is experiencing incredible market adoption and is rapidly dwarfing the sales of SAS/SATA flash drives.
The NVMe Working Group is also responsible for NVMe over Fabrics, which extends the NVMe standard to networks, enabling organizations to build networks that have latencies that rival direct attach storage. A key takeaway for IT professionals it that the NVMe specification is just beginning, the organization is continuing to push the capabilities of the specification to meet the needs of hyper-scalers, enterprises and consumers.
The latest NVMe Specification (1.4 ) features:
- Rebuild Assist, which simplifies data recovery and migration scenarios.
- Persistent Event Log, to enable robust drive history for issue triage and debug at scale.
- IO Determinism allows for better performance, isolation, and QoS. This feature alone could massively improve an organization’s flash experience.
- Multi-pathing enhancements or Asymmetric Namespace Access (ANA), which enable optimal and redundant paths to namespaces for high availability and full multi-controller scalability.
Each of these features is worth a blog of its own, which Storage Switzerland is going to work on with the NVMe Working Group, so mere IT mortals can understand the full ramifications of these new capabilities.
Briefing 3: Kaminario
Kaminario is an enterprise all-flash storage solution, designed for mission-critical enterprise class workloads. They were one of the first solutions to create a composable architecture and they are also one of the first companies to embark on the path of a pure subscription model.
Recently, Kaminario created a disaster recovery as a service (DRaaS) option that enables their install base to subscribe to a remote Kaminario system on a monthly basis as a disaster recovery target. The economics work well since the company doesn’t have to worry about maintaining a site and all the equipment in it.
Another recent addition is a version of K2, which enables the Kaminario software to run in AWS, Azure or Google Compute. Applications don’t have to be refactored, they run as is. The Kaminario solution can deliver better performance than the public cloud provider’s native block solution plus the customer can manage both cloud and their on-premises Kaminario environments from a single console. It also enables organizations to move applications and workloads easily, back and forth between the cloud and on-premises as necessary.
Briefing 4: Burlywood
Burlywood actually briefed us in detail last week, so this was a quick meet and greet. You can read our detailed summary of their FMS announcements in our briefing note “The Value of Fine Tuned Flash Drives – Burlywood Briefing Note”
Briefing 5: Formulus Black
Formulus Black also briefed us in advance of FMS but we sat down with them for a bit to discuss how the business is going. Formulus Black is an in-memory storage solution that enables any application to become an in-memory application. It manages all the challenges with DRAM based storage with a feature called BLINK. They also support Intel Optane Memory in its memory mode instead of the emulated storage mode. While BLINKs aren’t as necessary with Optane, the customer can still use them and benefit from Optane’s increased density versus traditional DRAM.
Read all the details in our briefing note: “Democratizing In-Memory Computing”
Briefing 6: Tachyum
On a different track is Tachyum, which is a processor technology on a single processor chip, that delivers industry leading performance for data center, AI, and HPC workloads. The company claims that the chip will outperform the fastest Xeon @ 10X lower power, with 3X lower sell price per MIPS (Millions of Instructions per Second) and 4X lower Data Center annual Total Cost of Ownership.
Of course, you don’t build a chip in a day (or week or year). It is a long process. Tachyum claims that they expect to provide FPGA emulation in Q1 of 2020 and be close to sending the processor to initial, sample set sized production by the end of 2020. The capabilities of the processor promise to give AI and HPC workloads new capabilities with lower cost and power consumption. Keep an eye on this one.
Briefing 7: Crossbar
We’ve covered Crossbar several times in their nine year history. Crossbar’s product is Resistive RAM (RRAM), which is a type of non-volatile memory. The company continues to position the product as a storage offering but is having greater success when focusing on specific use cases. One such use case is moving AI to the edge by combining their Crossbar Processing Unit (XPU).
An example might be a university that needs to do facial recognition. A university might have hundreds of cameras on campus and want to know who is visiting their campus. However, having all of these cameras report back to a central system to do the facial recognition consumes too much processing and isn’t fast enough. By building a solution where facial recognition is done in the camera, the university gets much faster results and won’t need as much camera to data center bandwidth.
We promised a busy day one, and there you have it. Day two is actually going to be busier. Stay tuned to Storage Switzerland for more details.