From: Brian J. Watson <Brian.J.W<atson@hp...> - 2004-01-17 11:44:06
Over three years in coming, the first release candidate for OpenSSI 1.0
is now available!! It features simple configuration of root filesystem
failover, improved installation tools with more automation, man pages
for almost everything, integration with Red Hat's 2.4.20-28.9 kernel,
plus much, much more. Try it out and let us know what you think!
As always, you can find it at OpenSSI.org.
Here are the release notes:
The last major feature added to OpenSSI before it turns 1.0 is simple
configuration of root failover for users who have shared disk hardware.
Setting this up is now an integrated part of the installation tools.
Furthermore, a new root failover node can be added without rebooting the
entire cluster. There is also support for failover of non-root filesystems.
Besides the fstab changes for shared disk filesystems, fstab management
has been enhanced so that you can specify filesystems/swap for any node
using the "node=" option.
Installation has been cleaned up and simplified. The buggy install
script from 0.9.96 has been replaced with an improved Perl script. Many
of the steps for installing OpenSSI and configuring the first node have
been automated. This includes building an OpenSSI ramdisk and creating a
grub.conf stanza. All user input is now validated, and the installation
tools avoid writing to disk until all user input has been gathered. Some
of the installation commands have been renamed, although old names are
available as symlinks for a little while.
Added long format arguments to some of the OpenSSI commands.
Man pages for almost everything have been added.
A localview command has been added, along with an option to run onnode
in localview mode. This mode limits the scope of visibility for
/proc/<pids> and /proc/sysvipc/* to the local node only, and it allows
you to create "local" sysvipc objects.
Another new feature is clusterwide /dev/pts support. Ptys are numbered
uniquely clusterwide, which makes utmp, who, etc. much happier.
Merged OpenSSI with Red Hat's 2.4.20-28.9 kernel.
Lustre client support has been re-instated in the kernel, although the
module is not yet being distributed with this release.
Increased node-monitoring frequency from 3 packets in 10 seconds to 100
packets in 10 seconds to decrease the likelihood of false nodedowns.
Introduced new libcluster APIs to support MPI.
Fixed a variety of bugs, including a "no USB keyboard" panic, a problem
where page-cleaning could hang indefinitely if a RAM based filesystem
had dirty pages, ssidevfs failover problems, a simultaneous boot
deadlock, initrd unmounting, flawed sig 0 semantics, and loads command
From: Brian J. Watson <Brian.J.W<atson@hp...> - 2004-01-19 13:15:53
Karl Vogel wrote:
> This might interest some people on this list... It appears that Matthew
> Dillon of DragonFly BSD fame has SSI plans for Dragon Fly.
That's a lot of work to do from scratch. Perhaps he'd like to port
OpenSSI to DragonFly BSD. :) That's still no trivial feat!