- Internet buffer bloat
Bufferbloat: Problem, Migitation, and Solution
VOIP and teleconferencing often perform much more poorly on today's Internet than the Internet of a decade ago, despite great gains in bandwidth. Lots of fiber, cheap memory, smart hardware, variability of wireless goodput, changes in web browser behaviour, changes in TCP implementations, and a focus on benchmarking Internet performance solely by bandwidth, and engineer's natural reluctance to drop packets have conspired to encourage papering over problems by adding buffers; each of which may introduce latency when filled.
The mistaken quest to never drop packets has destroyed interactivity under load, and often results in actual higher packet loss, as TCP's congestion avoidance algorithms have been defeated by these buffers. The lessons of the "RED manifesto" of 1997 have been forgotten or never learned by a new generation of engineers.
Bufferbloat mitigation by tuning queue length is beginning to be deployed in cable broadband systems. But solving bufferbloat requires careful queue management that must be present anywhere a queue may form. With the publication of the new CoDel AQM algorithm by Nichols and Jacobson (and inclusion in Linux 3.5) we now have the opportunity to solve rather than mitigate bufferbloat. I will also touch on fq_codel, which combines stochastic fair queuing and CoDel, and why we like the combination so much, and the remaining challenges.
Jim Gettys is at Alcatel-Lucent Bell Labs, USA. Today he works on bufferbloat in all of its forms, including helping establish bufferbloat.net to serve as a rallying point in its solution.
He was the Vice President of Software at the One Laptop per Child project, one of the original developers of the X Window System, and the editor of the HTTP/1.1 specification in the IETF. In 1997 he won Bob Metcalfe's Internet Plumber of the Year award on behalf of the group who worked on HTTP/1.1.
slides (PDF, 2MB)
The State of ZFS
Peter Baer Galvin
ZFS has taken the world by storm, and is still advancing. This talk will summarize the state of ZFS, including its availability, feature set, and recent changes.
Peter Baer Galvin is a seasoned tech writer, columnist, consultant, teacher and author. He is the CTO for systems integrator and VAR, Corporate Technologies (www.cptech.com). He's a Lecturer at Boston University and co-author of the Operating Systems Concepts textbooks. He's given talks and tutorials at USENIX conferences and other venues.
slides (PDF, 65MB)
Solid State Drives: Use, Performance, Caching, and More
Dan Noé, VeloBit
Solid State Drives can deliver high performance, but their prices still break the budget. SSD caching is a lower-cost method to improve application performance by taking advantage of fewer Solid State Drives to improve I/O. Dan will discuss SSD performance characteristics, best practices and risks of SSD deployment, as well as how SSD caching works and whether it can improve your performance.
slides (PDF, 1.5MB)
Solaris Dynamic Tracing - DTrace
DTrace is a revolutionary software framework that enables unprecedented observability up and down the entire software stack. DTrace was first introduced in Solaris 10, and continues to ship with Solaris, with ports to Mac OS X and FreeBSD. Others are underway. This talk will provide an overview of the DTrace framework and key components, as well as a tour of using DTrace to measure and observe system behavior.
slides (PDF, 1.2MB)