This shows you the differences between two versions of the page.
Both sides previous revision Previous revision Next revision | Previous revision | ||
realtime:events:rt-summit2016:jailhouse-kvm [2016/08/22 14:04] anna-maria delete information about wrong talk... |
realtime:events:rt-summit2016:jailhouse-kvm [2016/08/22 14:34] (current) anna-maria insert go back to schedule link |
||
---|---|---|---|
Line 5: | Line 5: | ||
---- | ---- | ||
- | ===== Abstract ===== | + | ===== Abstracts ===== |
- | The pthread condvar uses internal locking which is not configurable for PI. Using | + | ==== Lightweight real-time partitioning for Linux with Jailhouse ==== |
- | condvars in real-time applications can therefor lead to non-intuitive unbounded | + | The Jailhouse hypervisor project was created to partition multicore |
- | priority inversions, despite the use of a PI mutex. Patches to glibc have provided | + | systems with the help of virtualization technology - and Linux. It is |
- | non-POSIX (-np) extensions to the condvar implementation, allowing the condvar to be | + | designed for simplicity and, thus, provides isolation between partitions |
- | configured for PI. While these patches are broadly used with the PREEMPT_RT Linux | + | at minimal costs. |
- | kernel, they have not yet made it into the upstream glibc sources. | + | |
- | Concurrently, the glibc implementation of condvars was found to be non-compliant with | + | This talk will briefly present the concept of Jailhouse, its latest |
- | POSIX, which has lead to the development of a new condvar algorithm, which makes it | + | state of development and typical use cases around real-time as well as |
- | impossible to support PI using the existing operations. While a solution has not yet | + | safety critical systems. It will address the question which unique role |
- | been identified, there are possibilities involving modifying existing operation or | + | this project can play within the Linux ecosystem. |
- | introducing new PI futex operation which merit discussion. | + | |
- | ===== Audience ===== | + | ==== Real Time KVM ==== |
- | We are soliciting input on implementation from kernel developers with experience in | + | |
- | core locking mechanisms (futex, rt-mutex, etc.) as well as system library developers, | + | |
- | and real-time applications developers. The focus of the presentation is discussion of | + | |
- | potential paths to a solution which is POSIX compliant for the core and supports | + | |
- | Priority Inheritance. | + | |
- | ===== Biography ===== | + | KVM is now suitable for low latency real time workloads. |
+ | Getting there required several changes to the code, which | ||
+ | are now now upstream. Real time KVM also requires very | ||
+ | careful system configuration. This presentation will describe | ||
+ | some of the issues faced (and fixed) during development, and | ||
+ | pitfalls in deploying real time KVM. | ||
- | ==== Darren Hart, Intel Open Source Technology Center ==== | + | ===== Audience ===== |
+ | The talk addresses systems architects and developers who need to meet | ||
+ | real-time and/or safety requirements on highly integrated network, | ||
+ | entertainment or control systems that run Linux, or Linux aside other OSes. The talk addresses as well KVM developers and people looking to deploy Real Time KVM. There are equal | ||
+ | amounts of magic in the code, and in system configuration. | ||
- | Darren's focus is Intel Architecture in embedded, real-time, and safety critical | + | ===== Biography ===== |
- | applications. He is the Linux kernel maintainer for x86 platform drivers and represents | + | ==== Jan Kiszka, Siemens ==== |
- | Intel to the LF Real-Time Linux Project, and the OSADL SIL2LinuxMP project. He | + | Jan Kiszka is working as consultant and senior software engineer in the |
- | previously served as the MinnowBoard program architect for 4 years, providing technical | + | Competence Center for Embedded Linux at Siemens Corporate Technology. He |
- | technical direction to the original MinnowBoard, MAX, and Turbot boards. As the Yocto | + | is supporting Siemens sectors with adapting and enhancing open source as |
- | Project BSP technical lead, he redesigned the BSPs for Intel processors and wrote the | + | platform for their products. For customer projects and whenever his |
- | manual for Yocto Project kernel development. Prior to Intel, Darren led the Real-Time | + | spare time permits, he is contributing to open source projects, |
- | Linux development team at IBM for 6 years where he focused on real-time scheduling and | + | specifically in the area of real-time and virtualization. |
- | IPC mechanisms. | + | |
- | ==== Torvald Riegel, Red Hat ==== | + | ==== Rik van Riel, Red Hat ==== |
- | Torvald works for Red Hat's Platform Toolchain Team, focusing on | + | Rik van Riel is a principal software engineer at Red Hat, |
- | parallelism and concurrency. Recently, he has been revising the | + | and a long term contributor to the Linux kernel. He has |
- | implementations of most of the POSIX Threads synchronization data | + | contributed to the memory management subsystem, the scheduler, |
- | structures in glibc and contributed to the improvement of the futex | + | and various components related to virtualization. Rik is active |
- | manpages. He holds a PhD for his work on shared-memory synchronization. | + | in community projects like kernelnewbies.org and likes to hike |
+ | and rock climb in his spare time. | ||
+ | ---- | ||
+ | <WRAP rightalign> Go to [[realtime:events:rt-summit2016:schedule|Real-Time Summit 2016 Schedule]]</WRAP> |