Computational Storage

Move your algorithm to the data, not data to the algorithm

What is Computational Storage?

An architecture in which data is processed in close physical proximity to the storage device.

Primary benefit: reduce the amount of data that must move between the storage plane and the compute plane.

Traditional vs. Computational Storage

Traditional Architecture

The CPU processes data compute tasks such as compression; data is sent between storage and compute planes.

Computational Storage Architecture shifts data compute tasks to a hardware accelerator (FPGA), offloading the CPU. Data stays close to the compute, avoiding data movement on the slower CPU compute plane.

Coming Soon:
High-Speed 100 Gb/s NVMe Data Recorder

SNIA Terminology

CSS | CSP | CSD | CSA

Service, processor or drive? We’ve standardized on the SNIA definitions for these, which fall into the following categories:

CSS: Computational Storage Service

What is a CSS?

A data service or information service that performs computation on data where the service and the data are associated with a storage device.

It’s important to remember a CSS is not a device or module itself—rather it’s the acceleration services performed by a CSP or CSD. For example, the compression service that our 250-U2 can provide (through Eideticom’s NoLoad IP) is a CSS.

CSP: Computational Storage Processor

What is a CSP?

A component that provides Computational Storage Services for an associated storage system without providing persistent data storage.

As you can see in the diagram, the CSP is a separate device from FLASH persistent data storage. Acceleration services (CSS), such as compression, are offloaded from the CPU to the FPGA.

As a further advantage, the 250-U2 (a CSP module) transfers data to/from FLASH using peer-to-peer transfers, offloading the CPU from not only acceleration but data movement. This is enabled using Eideticom’s NoLoad IP.

CSD: Computational Storage Drive

What is a CSD?

A storage element that provides Computational Storage Services and persistent data storage.

CSDs provide very close computation and storage, with the tradeoff from CSPs in the loss of using traditional SSD drives in favor of integrated FLASH storage. An example of a CSD is the 250-HMS BittWare created for IBM; click here to read more.

CSA: Computational Storage Array

What is a CSA?

A collection of Computational Storage Devices, control software, and optional storage devices.
An example of a CSA is BittWare’s NVMe High-Speed Data Recorder reference designs.

Applications

We target applications when demand to process storage outpaces traditional architectures featuring CPUs.

Controller

Examples: Compression, Erasure Coding and De-duplication

Artificial Intelligence Inference

Big Data Analysis

Machine Learning

Content Delivery

Database Acceleration

Examples: RocksDB, Cassandra, Hadoop and MySQL

Flexible Form Factors

We have customization capabilities for building a range of current and emerging form factors like ESDFF. Talk to us about how we can turn your application needs into an enterprise-class solution!

PCIe Add-in Card (AIC)

U.2

M.2 Accelerator Module

EDSFF

Eideticom Investment

BittWare (a Molex company) is a strategic investor in Eideticom, a recognized thought leader in NVMe based Computational Storage solutions.

Their mission is to develop world-class Computational Storage Solutions for cloud and enterprise data centers. Eideticom’s NoLoad® Computational Storage Processor (CSP) is accelerating data center infrastructure, enabling greater scalability and dramatically lowering cost.

Los Alamos National Laboratory Collaboration

Eideticom and Los Alamos National Laboratory are collaborating on a storage acceleration solution using the BittWare 250-U2 computational storage processor. Key news from the press release:

  • World’s first NVMe-based computational storage compressed parallel filesystem built using Eideticom’s NoLoad® CSP and deployed in LANL’s Lustre/ZFS-based HPC parallel filesystems.
  • NoLoad CSP’s high performance compression engines provide scalable offload of storage centric services and enable capacity increases with no impact on performance.
  • NoLoad’s NVMe-compliant interface simplifies deployment of computational offload by making it straightforward to consume in servers of all types and across all major operating systems.

Got a Question?

Ask our technical staff where Computational Storage fits in your business.