With the explosion of data creation, storage, and analysis in virtually every industry, data centers and cloud service providers have evolved to meet the needs of today’s high-growth technologies. Adopting hyper-dense NAND storage such as QLC SSDs supports this evolution by delivering huge value to read-intensive workloads. However, datasets are also growing in write-intensive workloads—and these workloads can strain the endurance and performance of hyper-dense storage.
The solution? Cloud Storage Acceleration Layer (CSAL), a software tool designed to make adopting new storage technologies easy—especially for customers who may not have the resources to develop their own software to interface with a drive. Right now, in its early stages, CSAL is enabling QLC SSDs to store data more intelligently, with the goal of improving write performance and endurance overall.
CSAL provides a host-based Flash Translation Layer (FTL)1 in the Storage Performance Development Kit (SPDK)2 that optimizes drive endurance and write performance by intelligently managing the flow of I/O from the host. (See definitions and Figure 1 at end of article)
In this article, Dave Leone, Director of Data SSD Solutions Development at Solidigm—and one of CSAL’s founding fathers—talks about the origins of CSAL and where it’s making waves in the world of data storage.
Let’s take a closer look at how CSAL was created in order to shed more light on how it’s augmenting QLC performance and endurance.
As many know, the first Solidigm drives began as Intel drives—and it’s with Intel that CSAL got its start. As Manager of the Solutions and AE Team for Host Storage Software at Intel, Dave Leone was tasked with finding clients for Intel’s Optane system acceleration solution.
Soon, conversations began with Alibaba—the world’s biggest e-commerce company—looking to transition its Elastic Cloud Storage (ECS) from HDDs to NVMe SSDs. For this application, Alibaba was weighing two candidates: all-TLC storage, and QLC with Optane.
To sell Alibaba on the latter, Dave and his team (now a part of Solidigm), had to meet Alibaba’s specific performance demands. It quickly became clear that new software development was necessary. Luckily, the team was already well on its way to a solution: the earliest iteration of CSAL, spearheaded by Principal Engineers Wayne Gao and Mariusz Barczak, and Sr. Principal Engineer Kapil Karkra.
As Dave remembers, Alibaba wanted to “at least match hard drive performance—and, in fact they wanted to throttle some workloads at hard drive performance levels to ensure quality of service while maximizing capacity utilization at scale” with, of course, substantial savings on TCO (Total Cost of Ownership).
In a matter of months, his team was hard at work fine-tuning and applying CSAL. “We had 12 workloads from Alibaba, and we had to meet their latency and performance requirements for [each of] these different workloads,” says Dave. “We went through multiple iterations of negotiations about those targets and a lot of time spent working on the code to streamline I/O paths. Ultimately, we got there and Alibaba put CSAL into production at the beginning of this year.”
What started as an Optane solution has evolved to use Solidigm’s newly launched D7-P5810 SLC SSDs as a new Storage Class Memory solution for CSAL.
From concept to completion, the first implementation of CSAL took just nine months. The results were nothing short of remarkable: CSAL delivered 2X the density and 2X the performance versus Alibaba’s old HDD array.
Write-shaping for SSDs is only the beginning for CSAL. As Dave notes, “Our roadmap includes the integration of capabilities that will make adoption of data placement technologies like FDP (Flexible Data Placement) and ZNS (Zoned Namespaces) a breeze, as well as RAID capabilities to aid in redundancy and to double the performance of RAID5 write hole closure—and in the future, we will enable the use of CXL memory devices from SK hynix to store CSAL’s FTL metadata.”
Dave is also quick to point out that CSAL is completely open-source and other storage innovators are already working on expanding its functionality. For example, Solidigm partner StarWind has developed their own management UI and High Availability services for CSAL.
To enable further storage community innovation, Solidigm launched the open source CSAL Reference Storage Platform (RSP) at Flash Memory Summit 2023, a turn-key CSAL-based NVMeoF target implementation that can be deployed as a virtual machine or container that further encourages storage technology innovations. The CSAL RSP can be dropped into a production environment and immediately achieve the high performance and high density benefits that CSAL and Solidigm’s D5-P5336 deliver, all with minimal changes required to an existing host environment.
“We’re trying to build a community,” says Dave—and the benefits of this community are clear. Open-source improvements on CSAL through SPDK will lead to a wealth of performance benefits for existing Solidigm products, which will, in turn, lead to greater adoption for those same products.
For more information on CSAL and to download the CSAL Reference Storage Platform components, visit the Cloud Storage Acceleration Layer (CSAL) page on Solidigm.com
1) Flash Translation Layer (FTL): A software or firmware component used in SSDs to manage the interactions between the flash memory chips (NAND) and the operating system of a computer or device. A host-based FTL runs on the server’s central processing unit (CPU). This provides several advantages over other types of FTLs—drive-integrated FTLs and controller-based FTLs—including increased flexibility, customizability, interoperability, and user control.
2) Storage Performance Development Kit (SPDK): An open-source software project initiated by Intel that provides a collection of libraries and tools designed to help developers build high-performance and efficient storage solutions.