Developer Conference 2015

Unikernels are software appliances transformed into standalone kernels with all core ("kernel") services like scheduler, network stack, file systems etc. provided as libraries and linked directly into the application. The resulting binary is then run directly in a VM or bare-metal.

In this talk we'll inspect the three pioneers in the field, note their differences and explore what lies just on the brink of tomorrow:

http://www.openmirage.org/
http://osv.io/
http://rumpkernel.org/

Do you know which file system is faster? How they work, why sometimes use xfs. Are there cases when you need to ask which FS to use? Do you know how to measure the performance of your file system. I will give you the answers in my talk...

Performance Co-Pilot is a highly adaptable and established toolkit for those interested in examining the details of system performance. Designed from the ground up to compliment existing system monitoring, PCP is gentle in its approach as your statistics gathering mechanism. Likewise, PCP developers understand; what's most important for the sysadmin/end-user to monitor, may not be known, or even available to the upstream community. Hence, the following mantra is central to the PCP architecture: 'If it is important for monitoring system performance, and you can measure it, you can easily integrate it into the PCP framework'. This presentation will go over the basics of the toolkit, recent developments, how the developments fit within PCP's overall architecture, as well as some examples.

Performance Co-Pilot: http://www.pcp.io/

The Linux 'perf' tools continuously grows in scope as an observability toolchest, with more and more features being added.

In this talk the 'probe' and 'trace' tools will be showcased: the first allows adding dynamic probes in arbitrary points in the kernel and in userspace libraries and programs, collecting global and local variables, as well as callchains, that can then be used in conjunction with the other perf tools: record, top, trace, script, etc.

The later, trace, started as a super strace, one that allows stracing not just threads, but other targets such as the whole system, sets of CPUs, etc, in addition to threads. It also has a much lower overhead as it doesn't use the ptrace syscall. Support for tracing page faults and other events, as well as provide callchains leading to a syscall and other traced events are being planned and may even be ready by devconf'15 time!

An example of integration among perf tools, using 'probe' to allow capturing syscall arguments for pretty printing in 'trace', will be described, showcasing 'wannabe tracepoints', dynamic probes that may end up becoming real tracepoints.

How to use perf support in scripting languages. Explaning how perf interfaces scripting languages and showing practical examples of existing scripts and how to write your own.

Tinykdump is a minimal daemon to capture kernel-based crash dumping (kdump) memory image to usb storage. Compared to the traditional kdump solution, it is,

• more reliable
• has smaller memory foot-print
• more friendly to kernel developers

Enjoy!
The Tinykdump Team

Talk should focus on latest new feature in lvm2, how the user can use thin provisioning and caching together. How to reach better performance.