Installation guide

Release Notes
New in Scyld ClusterWare 4.8.1 Update - Scyld Release 481g0007
1. The base kernel is upgraded to 2.6.9-89.0.29 (481g0004). See https://rhn.redhat.com/errata/RHSA-2010-0718.html for
details.
2. The igb network driver has integrated an upstream fix to more frequently update /proc/net/dev statistics, which
means that beostat and IMF more accurately report network usage for chipsets that use that driver.
3. beostatus no longer requires that TORQUE be installed.
4. Fixes a bpcp -p bug where the mode was not properly preserved across the copy.
5. Fixes a Scyld ClusterWare ganglia bug where the network bytes/second data rates were being misreported.
New in Scyld ClusterWare 4.8.1 Update - Scyld Release 481g0006
1. The base kernel is upgraded to 2.6.9-89.0.28 (481g0003). See https://rhn.redhat.com/errata/RHSA-2010-0606.html for
details.
2. Scyld ClusterWare now includes an e1000e Ethernet driver, version 1.2.8 from http://sourceforge.net/projects/e1000/,
replacing the e1000e (version 0.5.18.3) that was introduced in CW4.3.1 as an improvement over the native RHEL4-U3
driver.
3. Scyld ClusterWare now includes an igb Ethernet driver, version 2.2.9, from http://sourceforge.net/projects/e1000/, re-
placing the igb (version 1.3.8.6) that was introduced in CW4.3.1 as an improvement over the native RHEL4-U3 driver.
4. Avoids the most common port number conflicts (beoservs beofs2/tcp port and BProc’s bproc port) by starting with
the default port numbers (possibly overridden by config file server directives), and flexibly incrementing these port
numbers as needed to find an available port. See the Section called Issues with port numbers for details.
5. Fixes a bpmaster daemon segfault that occasionally occurs when performing a concurrent reboot (e.g., /usr/sbin/bpctl
-S all -R) of a large number of nodes.
New in Scyld ClusterWare 4.8.1 Update - Scyld Release 481g0005
1. The base kernel is upgraded to 2.6.9-89.0.26. See https://rhn.redhat.com/errata/RHSA-2010-0474.html for details.
2. Fixes a problem that exhibits itself as a compute node needing an excessively long time to reboot (e.g., 15 minutes, vs.
the more common two minutes, approximately).
3. The cluster administrator may restrict compute node access to the master node, in much the same way as an admin can
assign access permissions to individual compute nodes. For example, /usr/sbin/bpctl -M -m 0110 disallows process
migrations from a compute node to the master, including migrations using bpsh and bpcp. Additionally, a new config
file keyword, nodeaccess, provides the ability to make these master node and compute node access restrictions persistent
across cluster reboots. See the config file comments and the Administrator’s Guide for details.
4. /usr/bin/bpcp -p now replicates the source file’s UID and GID for the target file. Previously, even when using the -p
option, the target file was owned by root.
5. /usr/bin/bpcp now guarantees that the target file exists when bpcp exits. Previously, bpcp may have exited with a
successful status before the target was created.
14