Presentations

High-throughput concurrent disk IO in FreeBSD (link), Adrian Chadd

There are many different methods for implementing disk IO heavy applications in user-space. These range from simple multi-process servers up through threaded applications with task queues, mmap(), sendfile() and POSIX AIO with the pros and cons of each discussed. Standard issues revolving around traditional disks and controllers will be covered along with the increasing shift toward cheap SATA and flash storage.

A closer look at the ZFS file system, Pawel Jakub Dawidek

ZFS is SUN's revolutionary file system. It merges roles of a file system and volume manager. The features list it provides is really astonishing and includes: simple administration, transactional semantics, end-to-end data integrity, cheap snapshots, storage pool model, immense scalability, no fsck and more.

ZFS is part of the FreeBSD system, the first FreeBSD release that includes ZFS is 7.0.

The author will focus on presenting ZFS design. He'll describe ZFS layers and features and explain how they work together. So prepare yourself for a journey into ZFS world - the most advanced file system ever made.

Pawel Jakub Dawidek is a FreeBSD committer.

In the FreeBSD project, he works mostly in the storage subsystems area (GEOM, file systems), security (disk encryption, opencrypto framework, IPsec, jails), but his code is also in many other parts of the system.

Pawel currently lives in Warsaw, Poland, running his company where he works on security related products.

The HAMMER File System (tgz), Matthew Dillon

HAMMER is a brand new BSD filesystem which saw first light in July 2008 with the DragonFlyBSD 2.0 release. This presentation will cover HAMMER's long list of features with a particular emphasis on its historical data retention and mirroring features, and future directions.

In a nutshell, the HAMMER filesystem is a B-Tree and extent-structured filesystem with fast (almost instant) crash recovery, a large 60 bit (1 Exabyte) addressable space, automatic 30-60 second granular snapshots with full retention of historical state immediately accessible via filename extensions, queue-less master-to-multi-slave mirroring, and other features.

HAMMER represents a new way of thinking about what a filesystem should do. With today's storage greatly exceeding what most individuals or businesses actually need, and a major requirement for live access to backups, it is no longer appropriate to destroy data simply because someone does a 'rm', or modifies the contents of a file or directory structure. HAMMER presents a storage paradigm which seamlessly integrates the live view with fine-grained historical access, fast and efficient queue-less mirroring streams which have little effect on production filesystem performance, and the ability to manage a long- term data retention policy across both production and backup systems.

Matthew Dillon is a well known Open Source developer. He currently runs the DragonFly BSD project at www.dragonflybsd.org.

BSD versus GPL (link), Jason Dixon

BSD vs GPL is a sweeping epic, focused on the dichotomy between good and evil. It peers inside the hearts and minds of the creators of these movements and dissects their battle for world domination. No common documentary will dare to follow the path that BSD vs GPL blazes.

Jason Dixon is an unknown help desk technician from Maryland. He enjoys long walks in the park and the occassional hair tousling. The audience is cautioned against making direct eye contact.

Network Refactoring, or doing an oil change at 80 MPH, Michael W. Lucas

This is a discussion of things to look at as a network administrator when you take over a new network. For the last two years I've been responsible for integrating networks all across North and South America into one private MPLS network. Lessons learned, lessons not learned, and the technical and administrative problems you can expect in such situations.

Michael W Lucas is a network engineer and system administrator responsible for a network that stretches across the Western Hemisphere. He is the author of Absolute FreeBSD, Absolute OpenBSD, and assorted other computer-related books. He lives in Detroit, Michigan, with his wife and pet rats. He has promised to wear pants to this conference.

Design and Implementation of the Portable C Compiler (pdf), Anders Magnusson

PCC was the first wide-spread portable C Compiler, written by S.C. Johnson in the late 70's. It became the de-facto standard C compiler included in most Unix systems and was the base for many other compilers.

In 2001 the PCC sources were released by Caldera and became freely available. This presentation gives an overview of its design and the changes that I have made to make it conform to modern standards while still keeping it small, fast, simple and easy to port to new targets.

OpenBSD's Position Independent Executables (PIE) Implementation (link), Kurt Miller

OpenBSD has randomized the load addresses of shared libraries for many years. This helps prevent attacks that are described as return-to-libc attacks. However, programs are linked at fixed addresses which provides some optimizations for executables over shared libs. When a program is complied and linked to be position independent (e.g. Position Independent Executable/PIE) some of those optimizations are waived for the ability to load the program at a random address. In this session, Kurt will discuss OpenBSD's PIE implementation. It's impact on existing security mechanisms such as W^X on i386 and the various enhancements needed to the runtime linker, kernel and other system libs.

Kurt J. Miller is a Software Consultant and has been an open source advocate for many years. He has been involved as a commiter on projects such as OpenBSD, Apache Tomcat, and the BSD Java porting team. Kurt is most active in the OpenBSD project where he has helped improve OpenBSD's runtime linker (ld.so), userland pthreads library and has been active in porting large applications to OpenBSD just as Sun's JDK's, OpenOffice, Eclipse and help improve stability of the Mozilla.org applications.

Introduction to DNSSEC, Jeremy C. Reed

The presentation will introduce the DNS Security Extensions (DNSSEC-bis). It will cover the need, creation, use, troubleshooting, and problems of DNSSEC signed zones. The presentation will also introduce EDNS0, new resource records, DLV, and DNSSEC related tools. The examples will be shown using ISC BIND. Due to limited time, the lecture will be very introductory while still covering these many topics.

DNSSEC extends standard DNS to add resource records and algorithms to provide source authentication (not encryption) for DNS. It is backward compatible, so non-DNSSEC systems can still use the normal DNS data.

Presented by Jeremy C. Reed. Reed is a NetBSD and DragonFly committer, board member for BSD Certification Group, and an ISC employee.

Reconnoiter, Theo Schlossnagle

Great advances in operating system architecture, filesystem, security, etc. have been made in the last 10 years. Great advances in monitoring have not. Reconnoiter's primary goal is to change that and provide powerful tools to infrastructure operators to achieve success.

The UNIX philosophy surrounding BSD it is old and proven. Simple tools combined to provide powerful and maintainable function. The world is a lot bigger, and while this concept still applies, the interpretation must change. Monitoring is a deep and important aspect of production systems management; yet the way we have gone about monitoring services over the past decade has not evolved. Monitoring encompasses both assessing availability and functional performance (fault detection) as well as tracking resource utilization (for trending, capacity planning, and post-mortem analysis). When you have tens of thousands of services being monitored, the old interpretation of "simple tools" must be reinterpreted.

Reconnoiter (https://labs.omniti.com/trac/reconnoiter/) reinvents monitoring architecture by focusing heavily on decentralization of configuration and operation and centralization of policy and analysis. Each component follows the UNIX philosophy of "single purpose tool," but on a more macroscopic level allowing thousands of services to be monitored on inexpensive servers in each data center.

This presentation will be a review of the Reconnoiter monitoring system on it's core development platform OpenBSD. We'll discuss the what you should expect to see returned after investing in a monitoring system. We'll talk about how Reconnoiter is designed to meet the demands of "large" system deployments. Lastly we'll take a whirlwind tour though Reconnoiter in real life.

Porting PCC, Michael Shalayeff

Porting PCC requires very little work. A typical back end is under 100k of code. We will demonstrate typical design using a couple of real architectures as an example.

Detecting TCP regressions with tcpdiff (link), Mike Silbersack

Determining if a TCP stack is working correctly is hard. The tcpdiff project aims for a simpler goal: To automatically detect differences in TCP behavior between different versions of an operating system and display those differences in an easy to understand format. The value judgment of whether a certain change between version X and Y of a TCP stack is good or bad will be left to human eyes.

The initial version of tcpdiff will focus on how TCP stacks react to packet loss. Using Lawrence Stewart's DPD, a set of simple tcp connections will be made repeatedly. Each time the connection is made, packet loss will occur in a different predetermined pattern. Tcpdump logs for each of these connections will be saved to disk. The resulting logs will then be run through tcpdiff, which will normalize and compare the set of connections from each version of the OS being tested. The places where one version of the OS retransmitted in a different manner than the base OS will then be flagged so that a developer can see how behavior has changed.

When Hardware Is Wrong, or "They can Fix It In Software" (pdf), Jason L Wright

In the world of hardware design, time to market pressures dictate the speed with which products are designed and developed. Sometimes in the rush, bad decisions are made and the device driver author has to compensate for the resulting ill-conceived hardware. In this presentation, several pieces of modern hardware will be examined to show the impact of hardware design decisions on the resulting driver. Particular emphasis will be placed on the more egregious errors and the performance impact. The idea is to poke a bit of fun at the bad design decisions and to also have the audience walk away knowing a bit more about what goes into the device drivers upon which they depend.

Jason L. Wright has been involved with the OpenBSD project as a developer since 1998. In that time he has developed device drivers for just about every class of device: network cards, USB devices, frame buffers, disk drivers, hardware cryptography accelerators, and more. Also during that time he has seen his share of odd hardware design choices and the "they can fix it in software" attitude. He now works for the Idaho National Laboratory as a cyber security researcher examining critical infrastructure.

An introduction to the Automated Testing Framework (ATF) for NetBSD (pdf), Julio M. Merino Vidal

The Automated Testing Framework, or ATF for short, is a project that provides a set of libraries to ease in the development of C, C++ and shell-based test programs as well as a collection of utilities to run these tests automatically and summarize their results in nicely formatted reports. ATF is an independent project supported by The NetBSD Foundation and, as such, it runs on many Unix-like operating systems. However, its development originated in and is mainly focused on the NetBSD Operating System; after all, the main goal of the ATF project is to provide a modern testing framework for the NetBSD Operating System.

In this presentation, I will explain ATF in the context of NetBSD. We will see why NetBSD needed a renewed testing framework, why the existing alternatives did not suit the needs of the project, which are the main features and goals of ATF, how ATF works within NetBSD and, if time permits, a few code examples and little demonstration of automated executions.

Julio Merino has been officially involved with the NetBSD project since November 2002. Since then, he has worked on many areas of the system, ranging from the packaging system to the kernel and, along the process, realized that automated tests are an extremely important tool to have for the successful development of a project the size of NetBSD. He is the main developer of ATF and this has been his latest contribution to NetBSD. At the moment, he is enrolled in a Ph.D. program at the Computer Architecture department of the Universitat Politécnica de Catalunya at Spain and pursuing an engineering internship at Google, Inc.