This presentation surveys the challenges posed by 10GbE, and the solutions for it. It focuses on Generic Receive Offload (GRO), which is a technique analogous to TSO (TCP Segmentation Offload). 10GbE's primary challenge is the 1500-byte MTU which was set in an era where network speeds were many orders of magnitudes below today's standard. This creates a bottleneck as data has to be segmented into 1500-byte chunks. TSO overcomes the costs associated with segmentation for transmission, while GRO attempts to do the same for reception. The implementation of GRO will be examined, along with possible extensions of GRO, such as cooperation with multiqueue reception.  The audience would be anyone with interest in high-speed networking. Kernel devleopment experience is not required.

With mobile and handheld devices gaining popularity rapidly, power consumption by these devices becomes a critical selling point for gaining the lead in a competitive market. In our talk we discuss how software plays a critical role in keeping the device cool and how one can design software for power management. Our case study is based on one of the popular usage models namely, video playback where we have optimized the software for reducing CPU and platform power consumption. Software optimization techniques include reduction of interrupts memory usage and synchronization between threads with which we have reduced the power consumption by about 25%. This talk is aimed for software developers to write power-aware and efficient software. This material can be used as training for developers who are new to power-aware coding techniques and methodologies. As part of this talk, we discuss various tools that developers can use to determine the power-efficiency of their code.

The OASIS ODF Technical Committee with representatives from industry (Sun, Microsoft, IBM, Adobe, Corel, Red Flag China 2000, etc.) and individuals, stewards the standard. In 2009, the TC is completing ODF V1.2 with significant new functionality in the areas of Formula, Accessibility and Metadata. This talk will focus on the application implementation implications of ODF V1.2, for IT leaders and developers of desktop, and cloud solutions. Beyond the completion of ODF 1.2, complementary work has progressed to support ODF interoperability and conformance needs. A new ODFDOM API and tools are available from the ODF Toolki Union, enabling the development of excitig new document-centric solutions. The talk is introductory in nature.

XCPU is a novel process management system that represents its interface as a set of files. It supports both diskless and nodes with disks as well as heterogeneous clusters. Because its state is distributed on all the nodes, XCPU can survive crash of the head node and supports the usage of multiple head nodes simultaneously. Unlike other services that depend on network file systems to deliver the executables to the nodes, XCPU implements a push model that scales better. Additionally XCPU uses tree-spawn method to avoid network bottleneck on job startup and increase the performance. The targeted audience is systems developers and administrators interested in using Linux for High Performance Computing.

Virtualization has taken a strong foothold in the Linux space, in particular in the enterprise market and the virtual server hosting space. Until now, virtualization has primarily been used on smaller PCs, however with Nehalem larger core counts and real NUMA is a reality, even in smaller systems.This talk will discuss the work being done to break the current limits of KVM and QEMU, in order to support much larger guests in terms of core counts, NUMA handling and number of I/O devices; ideally to run guests with hundreds of virtual CPUs. This talk will also look at how making the hypervisor more NUMA aware impacts migration and dynamic resource management. In addition, it will look at new usage models and users of Virtualization, such as in the high performance computing space. People attending this talk are expected to have good kernel and system knowledge.

The Linux Kernel is developing at a rapid pace. More and more features are added to the kernel. This leads to growth of the kernel binary (kernel bloat) but also to an increased cache footprint as well as higher complexity in critical code paths. The common experience is that the kernel becomes slower as time progresses. Faster hardware is needed to offset that effect.   The increase in cache footprint and complexity also leads to OS monitoring tasks such as the execution of the timer interrupt to become more invasive. The time the processor is taken away from the application during a scheduling cycle grows and so does the CPU cache use of the OS. This causes disturbances in the execution path of applications. The effects can be drastic for low latency dependent applications.

I am an active maintainer for a number of packages that are part of the GNU project, including GNU Common C , GNU ccScript, GNU ccRTP, and GNU ccAudio, as well as the GNU telephony application server, GNU Bayonne, and GNU SIP Witch call server. I have served as the voluntary chairman of the FSF's DotGNU steering committee (http://www.dotgnu.org), and as the communities elected representative to the International Softswitch Consortium.

KVM allows Linux to be a hypervisor, running many virtual machines on a single physical host.  In a modern virtualization environment, with high core count boxes, efficent cpu scheduling, and efficient I/O paths, memory can easily be the limiting factor to the number of virtual machines a single physical machine can host.  KSM provides a software mechanism for increasing virtual machine density without adding extra memory to the host.  KSM is code running in the Linux kernel scanning the memory of all the virtual machines running on a single host, looking for duplication and consolidating.  In fact, duplication is quite common since these virtual machines may be running identical or at least similar OS images.  With KSM we're able to improve virtual machine density by as much as 300% without impacting performance.  One of the great benefits of using Linux as the hypervisor means KSM is not limited to KVM and virtual machines, but can also reduce memory pressure with normal Linux applications. This presentation is technical and geared towards developers since it will focus on the kernel implementation of KSM.  A person attending this presentation can expect to learn how KSM works and in what deployment scenarios it will help.