If you a sysadmin, chances are high you’ve seen a very slow ps or non-working top. If you are a developer trying to parse /proc/pid/* files, you know it’s a mess. While developing and using OpenVZ and CRIU, we feel this pain too, so we found a way to cure it!
This talk presents a new Linux Kernel interface to collect information about processes, named task-diag. It discusses multiple shortcomings of the current interface, and shows how the new one avoids those.
In particular, task-diag employs binary format, and uses flexible means to specify which kinds of information and for which tasks is required. Modelled after sock-diag, it uses the request-response model and netlink message format. Tests of the new interface (using modified procps utilities) demonstrate up to 10x speedup. Other users of the new API, notably CRIU and perf, also proved it is fast and viable.
How fast is your website? What are its limits? In fact, what are the limits of every single bit of the technology infrastructure you own? One can make the case that knowing your limits is crucial to any successful IT project.
On the back of last year’s Christoph Lameter deep dive into raw performance, this presentation will try to uncover some lesser known ways of testing your infrastructure as a whole as well as providing orientation into the possible typical constraints it may have and how to overcome them. As an added bonus, there will be some emphasis on measuring by samples and scaling beyond Moore’s Law.
Daniel Fonseca is an Automation Lead Engineer at Rackspace, a new Role created within the company that further highlights the importance of automation within the business. He had been a Lead Engineer for 2 years and was awarded Best Newcomer of the Year 2015 for having created, in... Read More →
Monday August 22, 2016 4:20pm - 5:10pm EDT
Regatta
