OneFS: Understanding Data Compression in ISILON

Isilon F810 nodes allow you to perform inline data compression on your Isilon cluster. OneFS supports inline data compression on Isilon F810 node pools only. F810 nodes contain Network Interface Cards (NICs) that compress and decompress data received by the node. Hardware compression and decompression are performed in parallel across the 40 Gb Ethernet interfaces of F810 nodes as clients read and write data to the cluster. This distributed interface model allows compression to scale linearly across the all-flash F810 node pool as an Isilon cluster grows and additional F810 nodes are added.

You can enable inline data compression on a cluster that has the following:

• F810 node pool
• 40 Gb Ethernet backend network
• OneFS 8.1.3 or OneFS 8.2.1 or later releases

Mixed Clusters
In a mixed cluster containing node types other than the F810, files are only stored in a compressed form on F810 node pools. Data that is written or tiered to storage pools of other node types are uncompressed when it moves between pools.

At a high level, the benefits of Isilon s new compression functionality include:

• Simplicity: No configuration, default-on continuous compression eliminates the management burden.
• Efficiency: By using inline hardware offloading, Isilon minimizes any performance impact while maximizing storage efficiency.
• Transparency: Compression is natively integrated into the OneFS file system, making it transparent to applications and workflows.
• Harmony: It works in concert with other Isilon storage efficiency tools.

OneFS inline compression uses a lossless algorithm which reduces the physical size of data when it is written to disk. The compressed data is decompressed when it is read back. Compression is transparent to applications that sit on top of the file system including protocol-based services like Network File System (NFS) and Server Message Block (SMB).

The primary purpose of Isilon inline compression is to reduce the storage requirements for data. Inline compression usually has a smaller storage footprint, reduced power and cooling requirements, and a reduction in the overall per-TB storage cost. However, compression also helps to reduce the total amount of data written to storage devices. This is beneficial for solid state drives (SSDs) and other media with finite overwrite limits, by reducing the drive wear rates.

So what is under the hood?

OneFS inline data compression is consist of five principle components:

1. Compression Platform
2. Compression Engine
3. Compression Chunk Map
4. Compression Alerting and Reporting Framework
5. Compression Control Path

In this article, we focus on the platform and compression engine.

Inline data compression is supported exclusively on the new Isilon F810 all-flash platform, in which each node includes a compression offload adapter. This adapter provides look-aside compression to offload certain tasks from the network processor. This means that compression and decompression are transparently performed using a dedicated Field Programmable Gate Array (FPGA) with minimal latency, reducing CPU and memory resources.

The new F810 chassis is available with the following storage options:

In addition to dual-port 40 Gb Ethernet interfaces, each compression offload adapter contains an FPGA chip dedicated to the compression and decompression of data. The two Ethernet ports on each adapter card in the backend PCI-e slot of the nodes are used for redundant backend network connectivity.

Hardware compression and decompression are performed in parallel across the 40GB Ethernet interfaces of the F810 nodes as clients read and write data to the cluster. This distributed interface model allows compression to scale linearly across an F810 node pool as an Isilon cluster grows and additional F810 nodes are added. OneFS versions 8.1.3 and beyond support from 4 to 144 F810 nodes or 36 chassis per cluster. The best practice recommendation is a maximum of 40 F810 nodes per cluster.

The F810 nodes use an FPGA-based hardware offload engine on the backend PCI-e network adapter to perform real-time compression and decompression of data. This occurs as files are written to and read from a node in the cluster using a connected client session.

In addition to the FPGA, the hardware compression engine uses a proprietary implementation of DEFLATE. This provides for the highest level of compression while incurring minimal to no performance penalty for highly compressible datasets.

The compression engine consists of three main components:

Since they reside on the same card, the compression engine shares PCI-e bandwidth with the node s backend Ethernet interfaces.

OneFS also provides a software implementation as fallback if the compression hardware fails. In a mixed cluster, the software implementation is use on non-F810 nodes that do not have hardware compression. Both hardware and software compression implementations are DEFLATE compatible.