Vault 2016

GlusterFS is a scalable network filesystem to create large, distributed storage solutions. Sakshi Bansal and Susant Palai will talk about the need for a distributed file system and give an introduction to GlusterFS. The talk will mainly focus on how GlusterFS manages distributions of files and directories through Distributed Hash Table (DHT). They will also review on some of the recently introduced key features to DHT that improve performance. Finally they will give a short demo on the working of GlusterFS.

Some block storage device standards are working on a “logical depop" function which allows a system to decommission a defective physical element, e.g. a disk head or an SSD die or channel, reformat the device and continue using it with a reduced capacity. Such feature can allow reduced operation costs (delayed device replacement) but has the drawback of data loss (data under the remaining valid physical elements) and device downtime during re-formatting.

Online logical depop is another proposed new feature allowing retaining the device valid data and eliminating the need for a re-format. The basic idea is to introduce new commands for the host to discover the ranges of LBAs impacted by a defective element. Using this information, the host can take actions when an element failure event is suspected or reported by the device: deallocate the LBAs served by the defective element and operate the device in a thin-provisioned mode, amputate the LBAs or truncate the device LBA range to restore operation as a fully provisioned device with a lower capacity.

The goal of this BoF is to discuss the usefulness of such features and gather feedback from different application and system point of views for drafting a standard minimizing the impact of this feature on existing systems as well as enabling enhanced functionality for file systems,  device mapper drivers (including logical volume manager) and application level distributed storage systems (key value stores, object stores).

GlusterFS is a scalable network filesystem that enables users to deploy large distributed storage solution using common off-the-shelf hardware. It provides high availability and redundancy through Automatic File Replication(AFR) and Erasure Coding(EC). In this presentation Anuradha Talur will be talking about improvements put forth in GlusterFS around AFR and EC to effectively tackle some of the long standing performance issues faced by users. These enhancements are targeted towards improving performance of small file workloads, and automatic repair/self-heal so that storage is not in degraded state for longer time. The talk also focus on improvements for scaling up self-heal/repair load when hardware can offer more iops, scaling self-heal load down during heavy traffic on storage from applications, and addressing the high consumption of CPU due to non systematic erasure coding.

GlusterFS is an open-source (mostly) POSIX compliant distributed filesystem originally written by Gluster Inc and now maintained by RedHat Inc. Here at Facebook it had humble beginings: a single rack of machines, serving only a single use-case. Over the next 4 years it grew to thousands of machines, hosting 10s of petabytes of data. This presentation is a story of how this transformation occurred and the things we did to make it happen.

I'll cover how we manage & automate 1000's of GlusterFS bricks as well as dive into some of the patches we've contributed to the community to make GlusterFS easier to manage, monitor and scale.

Glusterfs's new tiering feature helps assure that data is available and accessible at the correct performance level thus satisfying both cost and performance concerns. This presentation will explain the Glusterfs's automated tiering feature followed by a live demonstration, and provide performance benchmark results from different hardware configurations. Glusterfs tiering monitors and identifies the activity level of the data and automatically promotes active data to hot SSD based storage and demotes inactive data to cold SATA based storage. Hence, the data movement can happen in either direction, based on the access frequency. The demonstration from Quanta/QCT labs (commodity x86 hardware) will show how to implement a hot tier based on SSD drives (optimized for performance) and attach it to an Erasure coded Glusterfs volume that is based on SATA drives (optimized for cost/capacity).

Replication is an important aspect of any storage solution for backing up data and ensuring High Availability. This is non-trivial in case of distributed systems because achieving consistency and HA with only 2x replication is difficult as there is no notion of quorum (majority voting). But not everyone can afford to store 3x copies just to achieve consensus in case of a disagreement. So they end up using 2x replication but face frequent split-brains due to flaky networks etc. - a common complaint on the gluster-users mailing list.

The presentation will focus on the arbiter configuration in glusterFS replica volumes and explain how it provides the same consistency guarantees as that of a full blown 3 way replica but without the 3x storage cost. It describes how the arbitration logic works to prevents files from ending up in split-brain, how to deploy and monitor them.

Traditionally Ceph has made use of local file systems like XFS or btrfs to store its data. However, the mismatch between the OSD's requirements and the POSIX interface provided by kernel file systems has a huge performance cost and requires a lot of complexity. BlueStore, an entirely new OSD storage backend, utilizes block devices directly, doubling performance for most workloads. This talk will cover the motivation a new backend, the design and implementation, the improved performance on HDDs, SSDs, and NVMe, and discuss some of the thornier issues we had to overcome when replacing tried and true kernel file systems with entirely new code running in userspace.

Over the past decade object stores have become more and more prevalent. They provide infinitely addressable storage but cannot be accessed by millions of lines of currently written IO code. The 2-Tier project is integrating the concepts of distributed OrangeFS burst buffer storage optimized for flash with the ability to tier data seamlessly to an object store. This talk will give an overview of the project, its status and provide attendees a peek into recent technology demonstrations on premise and in the AWS cloud with CloudyCluster.

This BOF session will provide an introduction to the NVMe over Fabrics protocol and the Linux reference implementation from some of the authors of the specification and the Linux implementation.

This BOF session will provide an introduction to the NVMe over Fabrics protocol and the Linux reference implementation from some of the authors of the specification and the Linux implementation.