A key trend in the ever-changing computing world is to bring everything necessary (processing, storage, analysis, networking) closer to where the data resides. The increasing amounts of data being generated and the number of datasets being created – and the valuable business insights captured in it all – are putting pressure on compute, storage and networking vendors to find ways to reduce the time necessary and the costs involved in the treatment. and analyze this data.
“This is an ongoing evolution in the industry, from a traditional, very compute-centric system, where everything goes through the processor and all the computational tasks there are handled in this central CPU chipset. and that DRAM build,” JB Baker, vice president of marketing at computing storage vendor ScaleFlux, says The next platform. “Over time, what we’ve seen is that a few things are happening. First, domain-specific computing is able to handle certain tasks much more efficiently than CPUs. For example, GPUs and TPUs for AI workload acceleration, and now SmartNICs have come out to offload much of the network processing that can be handled much better in domain-specific silicon than in a CPU.
Computer storage “is the next wave in handling tasks that aren’t handled well in a CPU, like compression, which can be extremely cumbersome if you’re trying to run a high compression level algorithm like a file format gzip, or things that are going to have to happen to most of the mass of data which, by doing those things in the drive, such as filtering a database, we can reduce the amount of data that had to go moving across the PCIe bus and polluting the DRAM or even across the network because networks are another bottleneck,” says Baker. “It’s really this evolution of domain-specific computing and the mass of data and data speed that creates all these bottlenecks in your compute infrastructure.”
As we wrote, computing storage – also known as “in-place processing” – is an emerging space driven by startups in a storage segment that has solved capacity and speed issues over the last decade, but which is still faced with cost and time. involved in moving it to where it can be treated. The idea is basically to process the data close to where it is already stored.
ScaleFlux is one such startup, launched in 2014 and which has since raised $58.6 million in six funding rounds, the most recent being last year when the company raised over $21.5 million. dollars, according to Crunchbase. Customers include Nvidia, Tencent and Alibaba and technology partners include Hewlett Packard Enterprise, Dell EMC, Lenovo, Inspur and Super Micro.
The publisher launched two generations of products, the first in 2018 and the second two years later. ScaleFlux Compute Storage Disks (CSDs), which bring together PCIe SSDs and embedded compute engines, which offload data-intensive tasks from compute to free up CPUs and reduce I/O bottlenecks that occur when sending huge amounts of data from storage devices to data center systems or disaggregated compute nodes.
Compute engines take advantage of Field Programmable Gate Arrays (FPGAs) and drives can be installed in standard slots for 2.5-inch U.2 drives or half-height, mid-range PCIe expansion cards. -length. IT admins can also tweak the settings to better balance application storage space, performance, latency, and endurance.
“Our primary goal is to have drives that are better than your regular NVMe SSD,” says Baker. “We’ve embraced the medical mantra of doing no harm – make sure we deliver the performance, latency, endurance, reliability, all the features you expect from your enterprise NVMe SSD, then add power. innovation to bring you further improvements to your system efficiency and drive efficiency.”
Now, the ScaleFlux is releasing its third generation of products this week, which are now in beta and will be available in spring 2022. The latest generation not only comes with improved computer storage systems, but also Arm-based silicon and a suite of discrete software aimed at making it easier for businesses to use computer storage.
“CSware is an integration of software and applications that takes into account the transparent compression capability of disks and leverages it to improve performance, reduce CPU load, and also reduce DRAM pollution,” says Baker.
There is software RAID in the CSware offering that will take advantage of disk compression to deliver RAID 5-like storage costs and RAID 10-like performance, he says. There is a compression-optimized KV store which Baker says is an alternative to using RocksDB as a storage engine for a database, halving the amount of CPU and RAM consumed and doubling the performance of write database.
A third piece of software can be used in relational databases such as MySQL and MariaDB to manage data updates, improve the write performance of these databases, and reduce disk wear.
ScaleFlux will likely offer a free version of the software and donate it to the open source community to help drive adoption of computing storage.
The company’s new SFX 3000-series storage processor and the firmware it contains are the computing power behind the vendor’s latest systems. The CSD 3000 series NVMe drive is the latest generation of the company’s flagship offering. It enables businesses to reduce data storage costs by up to three times and double application performance. It also increases the endurance of flash memory by nine times compared to other drives on the market, says Baker.
With four programmable Arm cores, organizations can deploy application-specific distributed computing functions in the storage device.
“In the CSD 2000 generation that is released today, we have the capacity multiplier capability,” he says. “We have excellent performance in storage. We do not have encryption and we are not using the NVMe driver yet. We have a ScaleFlux driver. Within 3000 [generation], we are moving to NVMe drivers to very simply support a wider range of operating systems and expand the market for these drives much more widely.
NSD 3000 Series NVMe SSDs are essentially a smarter offering than other devices, with twice the endurance and random-write and mixed-read/write performance of competitive NVMe drives. It is aimed at companies that may not need to adopt all the features of the CSD 3000 and are looking for a cheaper alternative.
“For those who aren’t sure exactly how they would consume the expanded capacity or how they would consume those cores, here’s a product where we turn those things off and it’s just a better NVMe SSD,” he says. “In the NSD, we keep the transparent compression-decompression function, but we just leverage it to improve endurance and performance and don’t give customers the keys to perform capacity multiplication.”
The CSP 3000 provides an integrated flashless compute storage processor and gives organizations expanded access to programmable cores so they can add additional features specific to their deployments. This offer can be used when businesses “have HDDs and they’re not looking to swap out the HDDs wholesale, but they want to do a compute offload to go with that array or for some other reason,” Baker says. “If they need to separate compute from storage, then CSP is a viable solution.”
Developing the Arm-based ASIC was key to allowing ScaleFlux to expand the capabilities of its portfolio, which couldn’t be done with an FPGA in terms of cost and power, he says. The system on chip (SoC) integrates four functional domains. There is a flash subsystem with a 16-channel controller and built-in compression/decompression and encryption. Also integrated are hardware compute engines for fixed algorithm tasks and a CPU complex for use-case-specific programs. In the CSD, users have access to four of the eight cores; the others are used for flash handling and exception handling. However, in the CSP they have access to all eight cores before there is no flash involved.
There’s also a PCIe interface, with eight PCIe 4.0 ports, although in most form factors there will only be four lanes used.
“It’s all integrated into a single chip,” he says. “We don’t need a separate processor to perform the computational functions and a discrete flash controller. Everything is integrated to save board space, improve performance, save power [and] allow for greater disk capacity.
This data-driven architecture offers a range of advantages over traditional processor-centric compute infrastructure, including the ability to move more of those compute functions to disks, Baker says. What ScaleFlux seeks to do is offer enterprises a way to embed even more of these computational tasks on their hardware.