- Issues in using AFS in a clustering environment, including such
things as using a translator(nfs/smb?) as a shared cache for large
cluster machines.
- Ideas on possible alternatives for enhancing performance on large
compute clusters.
- Future plans of OpenAFS development
- Other people's problems & experiences
- Software installation when a distributed filesystem is available -
the fine line between local installs and network installs, how we and
other people are managing this, especially in terms of modern software
with all the shlib dependencies.
- Using Amanda, an open source, free, network backup system to back
up AFS file servers -- specifically OpenAFS Linux based fileservers.
- Kerberos 5 migration. We are particularly interested in the
experiences of people who have migrated especially if they have
discovered (and solved) problems not covered by the migration
documentation. We are also curious about the current state of
interoperability between kerberos 5 used for AFS and Windows 2000/XP
clients.
- Supporting users with laptops or home systems with tenuous
connections to our AFS servers. This leads to two issues: The AFS
client, typically OpenAFS for us, needs to be able to shutdown and
restart reliably without having to reboot the system. Are other folk
are concerned about this.
- Experiences with Windows NT/2000/XP clients.
- Experiences with Mac OS X clients.
- The need for certain directories to allow for disconnected operation
and resynchronization when AFS is available. The simplest way for AFS
to do this is force everything of interest into the cache, work on it
and then let the cache synchronize things when AFS returns. This is
far from ideal. It would be nice to have something like the Windows
2000 offline folder with explicit synchronization and the possibility
of having the same directory offline on multiple machines with
synchronization doing "the right thing". Maybe this isn't the role of
AFS, but maybe it is.
- Can AFS be modified to change the way it responds to an overfull
partition? Other enterprise level file systems must have similar
issues. Has the problem been solved elsewhere? Are such solutions
applicable to AFS?
- We are using a memory cache with the IBM Linux client. We are about
to switch to the OpenAFS Linux client. Is a mem cache supported? Is it
a good idea or a bad idea? If it is not supported (or well supported),
why not? Would a disk cache on a ramdisk be better, and if so, why?
- OpenAFS is changing rapidly. All the bugfixes seem to be going into
the latest release along with new features. Should it move to a stable
and development tree model where bug fixes go into both trees, but new
features only go into the development tree? This is the way many open
source projects are going now and it is better for the users but
harder on the developers.
- Kerberos 5 support. At ANL, we have a Kerberos infrastructure as
well as an AFS infrastructure. While we use them together and have
many tools that rely on both, having Kerberos 5 support directly in
AFS would solve a few problems. I would like to talk about any work
being done in this area.
- Disconnected usage. One of the more interesting ideas that I saw in
the Coda project at CMU was being able to use the Coda system while
disconnected from the network. I have seen bits about adding that
to OpenAFS on their website, and would like to see what has been
done so far and talk about what else could be added to make it more
useful.
- AFS backups. We're currently using the built-in AFS backup mechanism, but
are migrating to Veritas NetBackup. I'd like to see what systems people are
using in this area and talk about any gotchas that non-native utilities may
include.
- AFS infrastructure tools. I would be interested in discussing what tools or
applications the other participants use or wrote to ease administration of
AFS, or to use AFS to simplify other administration tasks. At ANL, we have
a system that uses Kerberos and files stored in AFS to allow commands to be
run in parallel on several machines simultaneously. This program was
originally written to aid in system patching, but has since been extended
to support other types of work.
- MR-AFS availability
- managing backups using TSM
- monitoring & proactive AFS administration
- recovery from DB problems
- thought on database replication (performance impacts)
- performance tuning -- is RX really the problem?
- Coupling PTS with directories
- Win2k kerberos & AFS kerberos -- windows afslog?
- Experiences running AFS on Win2k terminal servers
- Have any OpenAFS patches gone back into IBM AFS yet?
- How to merge(?) the two CellServDB files? How can conflicts be
resolved?
- What are the largest machines running OpenAFS on other sites
(e.g. Sun E10K or 6800s). How many CPUs? In other words: do we
need/want IBMs AFS any more?
-
-
Back to the LISA AFS Workshop main page.