WekaIO pulls some Matrix kung fu on SPEC file system benchmark

Like a bat out of parallel...

Matrix kung fu

Startup type WekaIO has apparently walked all over IBM's Spectrum Scale parallel file system with a doubled SPEC SFS2014 benchmark score for its Matrix software running on Supermicro servers.

The benchmark tests the performance of filers and we're looking at the number of software builds completed in a run and the overall response time (ORT).

WekaIO's Matrix parallel file system seemingly showed IBM's Spectrum Scale a clean pair of heels, scoring 1,200 builds against Spectrum Scale's 600.

System Builds ORT Date
WekaIO Matrix – Supermicro 1,200 1.02 March 2018
IBM Spectrum Scale – E8 600 0.69 January 2018
NetApp FAS 8200 FlexGroup 520 1.04 September 2017
WekaIO Matrix 500 3.06 July 2017
IBM Spectrum Scale – Cisco UCS – FlashSystem 900 240 1.32 September 2017

The previous top-scoring system was IBM's Spectrum Scale parallel file system using an E8 NVMe all-flash array. That setup had a D24 storage node with 24 x HGST N200 1.6TB dual-port NVMe SSDs and 16 Spectrum Scale client nodes.

The WekaIO configuration had 4 x Supermicro Big Twin servers – 4 x 2-socket CPU nodes in 2U chassis with 4 x NVMe SSDs/node. The WekaIO Matrix 3.1 file system used 64 x 1.2TB Micron 9100 NVMe SSDs, and there were 11 client nodes hooked up via a Mellanox 100GbitE MSN 2700 switch.

We see NVMe SSDs being used in both cases.

Here are the ranked build score results:

Spec_SFS_2014_bar_chart

And here the build scores overlaid with the ORT numbers:

Spec_SFS2014_2_AXIS_CHART

Spect SFS2014 builds and ORT values

A previous WekaIO Matrix system, using v3.0 software, scored 500 builds on this benchmark in July 2017. The latest result featured updated Matrix v3.1 software. Is that why it is so much faster?

The startup is one of a crowd of techies ostensibly making large performance gains and latency lowering file system tech moves.

WekaIO co-founder and CEO Liran Zvibel said: "The current results are higher than the previous ones for two reasons: previous results were run on AWS R3 instances that have local SATA devices, so they did not support our native NVMe-oF stack (the new AWS I3 instances do).

"Secondly, we are already the fastest file system available, as we are the only file system to be completely re-architected for NVMe and low-latency fabrics."

It rather looks like its file system software made the difference. HPE has recently started reselling the Matrix software, which is certified with its Apollo 2000 Gen10 and ProLiant DL360 Gen10 servers.

Zvibel told The Reg that his firm had more software development work in the pipeline: "Our architecture still has a lot of potential for performance optimisations to become even faster." ®




Biting the hand that feeds IT © 1998–2018