Fosdem O'Reilly
2005 Edition Free and Open Source Software Developer's European Meeting


GNU Hurd Developers' room Schedule

[ News ]


Saturday Feb 26th 2005
Interactions in a Multiserver Operating System: The Importance of a good RPC Framework Unlike in a monolithic kernel where all resource management and
abstraction is implemented in the kernel, in a multiserver operating
system, untrusted servers provide the mechanisms and the policy.
These separate components must interact in a secure fashion if the
system is to fully provide protection domains and avoid unintended
sharing. As such the remote procedure call, RPC, system must be
well designed.

This talk will discuss the RPC system which has been designed for
the port to the Hurd on L4.

Marcus Brinkmann
The Device Driver Framework for multiserver operating systems running on L4 An introduction to the device driver framework for multiserver
operating systems running on the L4 microkernel.

Peter De Schrijver
Grub 2 GRUB 2 is a redesign and rewrite of the original GRUB, GRUB Legacy.
GRUB 2 will be important to GNU/Hurd in the future, just like a
portable multiboot standard. It is important to have the same
bootloader on all architectures.

This talk is about the advantages of the new GRUB. Things that will
be discussed are modules, the interfaces (disks, filesystems, etc),
portability and what there will be implemented in the future.

Marco Gerards
Supporting Larger Filesystems The inability to work with filesystems larger than 2G was one of the
most annoying limitations of the Hurd. This talk will describe the
technical problems which limited filesystems to 2GB and present the
approach which has been used to over come it in ext2fs.

Ognyan Kulev
Hurd on L4: Towards Extensibility The most compelling features of the Hurd--increased flexibility and security--have been realized in the Hurd running on Mach. Yet performance is dismal. It is true that GNU Mach has been optimized for early 1990s hardware, however, would tuning offer a sufficient performance increase? Having moved all but the core of a monolithic kernel to user-space, Mach's resource schedulers (VMM, CPU and I/O) can only use pattern analysis to predict resource usage. Linux, for instance, has shown how tenuous pattern analysis is even with some application specific knowledge about resource usage from e.g. file systems and network stacks. The resource manager requires application specific knowledge if it is to manage resources well. Hence, we must move file systems and other heavy resource users back into the kernel, i.e. head back towards a monolithic kernel and forfeit the advantages of the Hurd, or move the resource managers into user-space. We are pursuing the latter approach in the Hurd on L4.
Neal H. Walfield
Sunday Feb 27th 2005


© FOSDEM 2003-2005 - powered by Argon7