This week’s guest episode was recorded live in Silicon Valley at the offices of WekaIO.  The company has developed a scale-out parallel file system called Matrix that was specifically designed to exploit NVMe storage and new fast networking.  Chris is onsite to talk with CEO and co-founder, Liran Zivbel.  Martin is dialled in remotely from the bowels of the Storage Unpacked offices.

The conversation covers how Matrix was developed to work with new media and at the same time address some of the issues seen in the use of parallel file systems such as managing high levels of throughput and small file content.  Building a scale-out file system is challenging, and Weka developed their own user space storage operating system to ensure the performance of NVMe could be fully exploited.  Liran discusses the SPEC SFS 2014 benchmark tests, which can be found here.  Matrix is available for AWS on i3 EC2 instances.  More details can be found here (PDF).

Elapsed Time: 00:27:27

Timeline

* 00:00:00 – Intros

* 00:01:00 – What is WekaIO Matrix?

* 00:02:20 – Why are parallel file systems hard to build?

* 00:04:30 – What is it that end users want from a parallel file system?

* 00:05:00 – How is Matrix packaged and delivered?

* 00:07:00 – A real-time userspace storage operating system.

* 00:08:00 – Native Linux and a native file system driver.

* 00:11:00 – Snap to object – store snapshots on an object store.

* 00:13:00 – How is data protected within Matrix?

* 00:16:00 – How does performance scale with additional hardware?

* 00:17:00 – How does Matrix deploy in public cloud?

* 00:18:00 – Matrix SPEC SFS benchmark results.

* 00:20:30 – What are the typical customer deployment models?

* 00:22:00 – Storage pets versus storage cattle.

* 00:24:00 – Consumption models.

 Copyright (c) 2016-2018 Storage Unpacked.  No reproduction or re-use without permission. Podcast Episode F7G4.