September 27, 2006
This article was contributed by Stacey Quandt
The 13th annual International Linux System Technology Conference, also
known as Linux Kongress, took place September 5 - 8 in Nürnberg,
Germany. As a technical Linux event Linux
Kongress is smaller in scale than the Ottawa Linux Symposium and
linux.conf.au. Still the conference sessions and tutorials included a
number of quality talks from familiar members of the Linux and open
source communities such as Heinz Mauelshagen, Lars Mueller, Theodore
Ts'o, Volker Lendecke, Alan Robertson, and Daniel Phillips.
A few of the talks stood out. One such talk was Felix von
Leitner's presentation titled "Benchmarking, round 2: I/O
Performance", in which he tested file system performance on
Linux, Windows, OpenSolaris, NetBSD, FreeBSD, and OpenBSD in order
to better understand the scalability of different operating systems and
IP stack throughput. Based on von Leitner's benchmarking methodology Linux
has the fastest file system - reiser4.
The testing theme continued with Poornima
Bangalore, whose presentation was on the topic of "Best Practices
in Linux Kernel Testing." Her talk detailed many of the key
differences between traditional and open source testing. She pointed out
that mainline kernel testing is more challenging than testing many other
open source projects because of the rapid development and the different
sub trees in the kernel: the stable kernels are released every 6 weeks or so,
release candidate (-rc) kernels are available every week, and experimental
(-mm) kernels are available every few days. Poornima shared best practices
regarding kernel configuration, hardware configuration, test automation,
test coverage, and first failure data capture.
Heinz Mauelshagen gave a talk on
device-mapper architecture features and the related target feature
set. In the talk "Linux as a Hypervisor," Jeff Dike discussed the
evolution of the hypervisor support in the Linux kernel and how
capabilities such as ptrace, AIO and O_DIRECT make a difference to
virtual machines. He also talked about the implications of FUSE
(filesystems in userspace) and the manageability benefits of exporting a
UML filesystem to the host. Lars Marowsky-Bree's presentation on
Heartbeat 2 and Xen
explored Heartbeat's ability to manage Xen
guests. He expanded on Heartbeat's architecture and its integration with
Xen to enable resource reallocation, globally ordered recovery actions,
and data center automation policies using the Cluster Resource
Manager (CRM).
Mattias Rechenburg's presentation on
"Using Enterprise Data
Centers with OpenQRM" showcased the state of
OpenQRM an open source project to achieve high-availability,
scalability, and deployment, service and server virtualization on a
variety of operation system. In spite of OpenQRM's pluggable architecture,
the audience focused on the fact that it depends on a binary module
which requires support from Qlusters. The general sentiment from the
audience was they were not interested if they couldn't get support from
Red Hat, IBM, Hewlett-Packard etc.
In "Real-Time Approaches to Linux,"
Ted Ts'o shared his perspective on enterprise real-time computing and
how it differs from so-called traditional real-time computing.
He emphasized the changing
requirements in enterprise software and how high throughput is not
enough because customers increasingly also require latency guarantees,
especially in particular military applications and trading
systems. It was interesting to hear about the benefits and tradeoffs of
different approaches to enterprise real-time including RTAI and Ingo
Molnar's CONFIG_PREEMPT_RT.
Ted suggested
that guidelines outlined by his colleague Paul McKenney can be used to
evaluate the different approaches to enterprise real-time. This includes
quality of service, the amount of code inspection required when a new
feature is added, the API provided to applications, the relative
complexity, fault isolation, and supported hardware and software
configurations.
Although IBM presently has only one customer that plans to
deploy enterprise real-time computing, the ability to support large
SMP systems,
TCP/IP, commercially available middleware, and databases makes it an
area to watch in the future. Ted also elaborated on the features of
IBM's real-time JVM/SDK (aka IBM Websphere Real-Time v1.0) such as RTSJ
(Real-time specification for Java), the Metronome real-time garbage
collector, and AOT (Ahead of Time Compilation). The talk emphasized that
there are many new applications for real-time operating systems, and in
particular enterprise real-time Linux.
Maddog provided the final keynote on having fun
with open source in his own inimitable way.
(
Log in to post comments)