Cephalocon: SoftIron Released Hardware-Based Ceph Erasure Coding Acceleration

SoftIron Ltd. launched dedicated hardware to accelerate erasure coding at Cephalocon, and revealed how it will alter the standard practice of data redundancy, saving enterprises millions in the process.

Codenamed Accepherator, the company’s hardware acceleration I/O module uses programmable silicon to compute erasure coding on the fly. It’s for the enterprise-storage industry in that it solves the traditional ‘3X replication’ data redundancy issue without the expense of high-end CPUs, whilst also sitting inline as a 10GbE network interface.

Data redundancy refers to the replication or backups of data sets that are required to ensure complete restoration of any lost data in the event of drive failures, and it forms a core part of any storage strategy. However, it’s also very expensive.

Phil Straw, CTO, SoftIron, explains: “Data redundancy in a Ceph environment is usually solved by creating three copies of data and spreading it across multiple drives via the CRUSH algorithm. The idea is that if a drive fails, and data is seemingly lost, there are still two additional copies of the data to recover data from. The challenge is that three times replication is expensive, you need to buy and maintain three times the useable storage you need in order to store the data with redundancy. As you can imagine, this approach adds a lot of expense to storage, so we’ve turned to erasure coding.“

“Erasure Coding (EC) works by processing data through an algorithm that breaks the data up into chunks and writes a single copy with extra parity bits which can be used to rebuild the data in the event of lost media. Traditionally, EC is also expensive o run because it requires a lot of CPU horsepower to run correctly, which essentially just shifts the cost of data parity and transfers it to the CPUs of the storage appliance. This additional CPU load slows storage writes and the only way to solve it has been to use the most expensive, power-hungry CPUs on the market.“

The company created a dual-purpose EC accelerator + 10GbE SFP network interface to solve this issue, and in so doing has reduced the costs and complexity of data redundancy. It does so without the expense of triple replication and without the CPU penalty of traditional erasure coding.

Offered as an I/O (NIC) option for HyperDrive – the company’s custom-built, dedicated Ceph appliance for software-defined storage – it works as an I/O module that computes erasure coding on the fly at line rate, removing the load from the CPU while also providing a 10GbE SFP interface.

“Now, with what we call Accepheration, we can confidently deliver reliable redundancy yet at a fraction of the traditional cost. We’ve managed to cut the overall TCO in half by virtue of the fact you now only require 1.33x replication, and you don’t need expensive CPU’s for the erasure coding. It’s a game-changer in the true sense of the expression,” says Straw.