Installation guide

Release Notes
6. The beostat service that supplies cluster performance statistics to Scyld IMF, beostatus, ganglia and other cluster status
visualization utilities now understands bonded network devices. Previously, network statistics were double-reported:
counting both the aggregated bonded pseudo-device and the individual devices that comprise the bonded device.
7. Eliminates a bogus recvstats syslog message of the form "Received stats from IP addr" that occasionally appeared as a
compute node starts up.
New in Scyld ClusterWare 4.8.1 Update - Scyld Release 481g0004
1. The base kernel is upgraded to 2.6.9-89.0.25. See https://rhn.redhat.com/errata/RHSA-2010-0394.html for details.
2. Fixes a bug where doing a ctrl-c or a kill of certain workloads might leave a "lingering ghost" process on the master node:
a process that was associated with the real process that had been executing on a compute node and which was properly
terminated by the ctrl-c or kill. Additionally, previously a "lingering ghost" process could not be manually killed, and it
would only get cleaned up when the cluster rebooted. Now "lingering ghosts" should not appear. If any does appear, it
can now be killed using /bin/kill or /usr/bin/killall, as appropriate.
3. Fixes an infrequent BProc bug which exhibited itself most commonly as a kernel panic due to a segfault in ghost_put or
to a "Kernel BUG at spinlock:119" called from bproc_purge_requests.
4. Eliminates a bogus BProc syslog message "proc.exe not null".
New in Scyld ClusterWare 4.8.1
1. The base kernel is upgraded to 2.6.9-89.0.23. See https://rhn.redhat.com/errata/RHSA-2009-1541.html,
https://rhn.redhat.com/errata/RHSA-2009-1671.html, https://rhn.redhat.com/errata/RHSA-2010-0020.html,
https://rhn.redhat.com/errata/RHSA-2010-0076.html, and https://rhn.redhat.com/errata/RHSA-2010-0146.html for
details.
2. Scyld ClusterWare now supports non-Scyld nodes as compute nodes in the cluster, in addition to the traditional Scyld
nodes that integrate into the Scyld unified process management environment. An example of a non-Scyld compute node
is a server that executes a full distribution of Red Hat Enterprise Linux (RHEL) or CentOS and which boots from a local
harddrive. See the Administrator’s Guide for details.
3. Supports two new /etc/beowulf/config keywords, host and hostrange. The config file may contain zero or more
of each. A host directive pairs a unique client MAC address with the unique IP address to be delivered to that client,
together with an optional name for the client, for use if and when that client makes a DHCP request to the master node.
A hostrange directive specifies a unique range of IP addresses that does not collide with the iprange addresses used for
cluster compute nodes, nor with the IP address(es) used for master node(s). Every host IP address must fall within one of
the hostrange ranges. These clients are typically some device or node on the cluster private network other than a compute
node, such as a managed switch or some other device that uses DHCP to obtain an IP address. See the Administrator’s
Guide for details.
4. Fixes a bug in BProc where certain workloads would cause a master node kernel panic, most commonly a segfault in the
routine ghost_put.
5. Fixes a bug in node startup which ignored a fatal mount failure and allowed the node to transition to the up state. Proper
behavior is to abort the node startup and to leave the node in error state.
6. Fixes a bug where certain workloads would generate many thousands of sockets sitting in TIME_WAIT limbo, which is
at best inefficient and at worst would lead to temporary socket exhaustion.
15