Installation guide

Release Notes
Optionally adjust the size limit for locked memory
OpenIB, MVAPICH, and MVAPICH2 require an override to the limit of how much memory can be locked.
Scyld ClusterWare adds a memlock override entry to /etc/security/limits.conf during a Scyld ClusterWare upgrade
(if the override entry does not already exist in that file), regardless of whether or not Infiniband is present in the cluster. The
new override line,
* - memlock unlimited
raises the limit to unlimited. If Infiniband is not present, then this new override line is unnecessary and may be deleted. If
Infiniband is present, we recommend leaving the new unlimited line in place. If you choose to experiment with a smaller
discrete value, then understand that Scyld ClusterWare MVAPICH requires a minimum of 16384 KBytes, which means
changing unlimited to 16384. If your new discrete value is too small, then MVAPICH reports a "CQ Creation" or "QP
Creation" error.
Optionally enable automatic CPU frequency management
If you wish to enable automatic CPU frequency management, you must have the base distribution’s kernel-utils package
installed, and then enable the Scyld ClusterWare script:
/sbin/beochkconfig 30cpuspeed on
You may optionally create a configuration file /etc/beowulf/conf.d/cpuspeed.conf (or node-specific
cpuspeed.conf.N), ostensibly derived from the master node’s /etc/cpuspeed.conf, to override default behavior. See
man cpuspeed for details.
Optionally enable SSHD on compute nodes
If you wish to allow users to /usr/bin/ssh or /usr/bin/scp from the master to a compute node, or from one compute node to
another compute node, then you must enable sshd on compute nodes by enabling the script:
/sbin/beochkconfig 81sshd on
See the Administrator’s Guide for details.
Optionally increase the ip_conntrack table size
Certain workloads doing IP forwarding may trigger a syslog message ip_conntrack: table full, dropping packet. If IP for-
warding is enabled, then at cluster startup time Scyld ClusterWare insures a max max table size of at least 524,288 and a
related table hashsize of 65,536 (maintaining the default 8-to-1 ratio for performance reasons). However, this max value
may still be inadequate for local workloads, and the table full, dropping packet syslog messages may still occur. Use:
/sbin/sysctl net.ipv4.ip_conntrack_max
to see the current max size, then keep manually increasing the max until the syslog messages stop occurring, e.g., use:
/sbin/sysctl -w net.ipv4.ip_conntrack_max=Nmax
to try new Nmax values. An effective Nmax also determines an effective Nhash hashsize: 1/8th the Nmax value. Make these
values persist across master node reboots by adding:
8