Installation guide

Release Notes

If yum fails with a Transaction Check Error that complains that a base distribution rpm is newer than the Scyld Cluster-

Ware rpm that is attempting to replace it, then you must manually install the downlevel Scyld ClusterWare rpm(s). For

example, if the complaint is about the kernel rpms, then do:

cd /var/cache/yum/

ls cw-*/packages/kernel-*

and locate the newest Scyld ClusterWare 2.6.9-100.490g0000.ELsmp kernel, which should reside in either

cw-core/packages/ or cw-updates/packages/. Then install that newest kernel:

rpm -iv --oldpackage cw-*/packages/kernel-smp-*490g0000*rpm

and then repeat the groupupdate, which should now Complete! successfully.

8. If the previous yum groupupdate excluded the openmpi-scyld* packages, then you can explicit download new packages,

if any are available, and install them in a way that avoids removing older openmpi-scyld* packages:

rm -fr /tmp/ompinew

mkdir -p /tmp/ompinew

yumdownloader --destdir /tmp/ompinew openmpi-scyld*

rpm -iv /tmp/ompinew/openmpi-scyld*

Note the use of yumdownloader and rpm -i. This is necessary because doing yum install openmpi-scyld* would

not, in fact, do a simple install and retain older packages. Rather, it would perform an update and remove the older

openmpi-scyld* package(s).

9. If the cluster uses Panasas storage, then you should have already downloaded the Panasas rpm that matches the Scyld

ClusterWare 4.9.0 kernel you have just installed. Now install the Panasas rpm using rpm -i.

10. Compare /etc/beowulf/config, which remains untouched by the Scyld ClusterWare update, with the new

config.rpmnew (if that ﬁle exists), examine the differences:

cd /etc/beowulf

diff config config.rpmnew

and carefully merge the config.rpmnew differences into /etc/beowulf/config. Please see the Section called Re-

solve *.rpmnew and *.rpmsave conﬁguration ﬁle differences for details.

Similarly, the preexisting /etc/beowulf/fstab may have been saved as fstab.rpmsave if it was locally modiﬁed.

If so, merge those local changes back into /etc/beowulf/fstab.

11. Disable (or delete) any old /etc/beowulf/init.d/ scripts that may be lingering from older Scyld ClusterWare re-

leases.

/sbin/beochkconfig --list

displays the list of existing Beowulf scripts. Common old scripts are ipmi (replaced by 20ipmi), torque (replaced by

90torque), and unnumbered inﬁniband scripts (replaced by 15openib and 16ipoib). memdev is no longer necessary

and should be deleted. Any such old script should be disabled (e.g., /sbin/beochkconﬁg ipmi off) or deleted (e.g.,

/sbin/beochkconﬁg --del ipmi).

12. Examine /etc/grub.conf to conﬁrm that the new 2.6.9-100.490g0000.ELsmp kernel is the default, then reboot your

master node.

13. If upgrading from Scyld ClusterWare 4.2.x, the ﬁrst time Beowulf services start, e.g., when doing /sbin/service beowulf

start or /etc/init.d/beowulf start, you will be prompted to accept a Scyld ClusterWare End User License Agreement

(EULA). See step 11 in the Section called Installing Scyld ClusterWare 4.9.0 on a Non-Scyld ClusterWare System for

details.