Vault 2016 has ended

Sign up or log in to bookmark your favorites and sync them to your phone or calendar.

Dev Focus [clear filter]
Wednesday, April 20

11:30am PDT

Solving the Linux Storage Scalability Bottlenecks - Jens Axboe, Facebook
Flash devices introduced a sudden shift in the performance profile of direct attached storage. With IOPS rates orders of magnitude higher than rotating storage, it became clear that Linux needed a re-design of its storage stack to properly support and get the most out of these new devices.

This talk will detail the architecture of blk-mq, the redesign of the core of the Linux storage stack, and the later set of changes made to adapt the SCSI stack to this new queuing model. Early results of running Facebook infrastructure production workloads on top of the new stack will also be shared.

avatar for Jens Axboe

Jens Axboe

Jens Axboe is Software Engineer at Facebook, formerly a Fellow at Fusion-io, and Consulting member of staff at Oracle. He also serves as the Linux block layer maintainer. Jens has worked on all things Linux IO related, such as data writeback, IO scheduling, SATA/SCSI, and others... Read More →

Wednesday April 20, 2016 11:30am - 12:20pm PDT
State Ballroom B

2:00pm PDT

A Small Case Study: Lessons Learned at Facebook - Chris Mason, Facebook
Facebook uses Linux everywhere, and our wide variety of workloads often expose problems in the kernel's handling of storage and memory management.

This talk will focus on issues we've encountered, how we fix them in production, and the projects we have underway to improve the kernel.

avatar for Chris Mason

Chris Mason

Software Engineer, Facebook
Chris is a Software Engineer on the kernel team at Facebook, and the maintainer of the Btrfs filesystem. He has been working full time on the kernel for over 16 years, and lives in Rochester New York.

Wednesday April 20, 2016 2:00pm - 2:50pm PDT
State Ballroom B
Thursday, April 21

10:30am PDT

Filesystem Fuzzing with American Fuzzy Lop (AFL) - Vegard Nossum & Quentin Casasnovas, Oracle
American Fuzzy Lop (AFL) is an open source fuzzing framework that relies on code instrumentation of a target program to find inputs that will cause the program to take new (and hopefully interesting) code paths. We have applied AFL to a range of Linux filesystem drivers and have quickly found a multitude of new bugs not found with regular ("dumb") fuzzers. Additionally, the testcases found by AFL can be used as a regression test suite that will help increase the confidence that any future change to the filesystem driver does not accidentally introduce bugs.

Through our presentation we share our techniques and tools directly with filesystem developers and make filesystem fuzzing with AFL more accessible to a wider audience. In this way, bugs can be found (and fixed) faster by the people who already know the filesystem code intimately.


Quentin Casasnovas

Senior SW engineer, Oracle
Quentin graduated from EPITA, a french engineering school, in 2010.He's started working as an embedded engineer for MathEmbedded, a Britishstart-up, then as a freelancer for Intel.  He's now working for Oracle inthe Ksplice tight knit team where he prepares Ksplice updates and try... Read More →

Vegard Nossum

Senior Developer, Oracle
Vegard graduated from the University of Oslo in 2012 with a thesis on the topic of SAT solving and has been working for Oracle on Ksplice kernel updates and infrastructure ever since. His experience with the Linux kernel includes writing kmemcheck, a tool for detecting use of uninitialized... Read More →

Thursday April 21, 2016 10:30am - 11:20am PDT
State Ballroom B

11:30am PDT

Scaling the Btrfs Free Space Cache - Omar Sandoval, Facebook
Btrfs is being deployed in more and more production systems, and as a result it has matured in stability and performance. In particular, Facebook's production systems push Btrfs especially hard, which has uncovered some unanticipated scalability bottlenecks. In this presentation, Omar Sandoval will discuss the shortcomings of the original Btrfs free space cache and explain the design of its replacement, the free space tree, which is Btrfs' latest solution to the classic filesystem problem of tracking unallocated space.


Omar Sandoval

Omar Sandoval is employed by Facebook where he works on the Linux kernel team. He has made numerous contributions to Btrfs, the most notable being the free space tree.

Thursday April 21, 2016 11:30am - 12:20pm PDT
State Ballroom B

2:00pm PDT

Understanding Client Side Shared Cache with Pblcacle - Luis Pabon, Red Hat
Cloud data centers deploy a large number of compute nodes to meet increasing application needs. The computes nodes introduce storage cluster contention, resulting in increased application response time. At last year's Linux Vault, we introduced Pblcache[1], a persistent, write-through, user space application block cache based on NetApp’s Mercury paper. It was demonstrated that a it not only reduced application response time, but also dramatically increases the IOPS capacity of a backend storage system. In this presentation we discuss the benefits and disadvantages of a shared client side block cache architecture based on Pblcache. We also discuss the possible methods of integrating Pblcache with Ceph libRBD and QEMU.

[1] https://github.com/pblcache/pblcache

avatar for Luis Pabon

Luis Pabon

Principal Software Engineer, CoreOS
Luis Pabón is a software engineer at CoreOS. Prior to joining CoreOS in November of 2016, he worked at Red Hat Storage, NetApp Advanced Technology Group, and at EMC on various storage products. He also previously presented at Vault in 2016 and 2015.

Thursday April 21, 2016 2:00pm - 2:50pm PDT
State Ballroom B
Filter sessions
Apply filters to sessions.