Rembrandt cluster installation

New kernel

We need to synchronize the kernels on the Rembrandt cluster to at most three kernels, available on all nodes:

• Default 2.6.8-10-amd64-k8-smp (Debian package kernel-image-2.6-amd64-k8-smp)
• Web-100 kernel
• Kernel with support for Chelsio NIC

Ideally, we like to combine the last two, and use that as the kernel on all the nodes. However, the Chelsio driver must patch the kernel, and at this moment the patch only works with the 2.6.6 kernel. On the other hand, there is no web-100 patch for 2.6.6.

Order of disks and network interfaces

The disadvantage of using different kernels on a cluster is that the order of the disks and the network interfaces are different.

When using the Chelsio-enabled kernel, with Chelsio NIC in the computer, that NIC will be eth0, while in the other nodes eth0 is the management interface. This forces us to have different /etc/network/interfaces files. This problem is described in detail on the LogicalInterfaceNames.

In the default kernel, support for the 3ware 9500 SATA controller resides in a loadable module, and the SATA RAID disk is available under /dev/sdb. However, in the current web-100 kernel, support is built into the kernel, and the SATA RAID disk is available under /dev/dsa, while the system disk is /dev/sdb. This forces us to change /etc/fstab and the kernel paramaters (in /boot/grub/menu.lst).

RAID configuration

[5 April 2005] The RAID disk is configured on rembrandt1-4; on rembrandt5-8 the 3ware driver is still initializing it.

For rembrandt5-8 we still need to:

• BIOS: wijzig boot disk
• eerste boot: kernel params: wijzig sda in sdb & disable /dev/md0 (optional)
• /boot/grub/menu.lst: wijzig sda in sdb (optional)
• apt-get install xfsprogs
• mkfs -t xfs /dev/sda (or /dev/sdb)
• /etc/mdadm/mdadm.conf: weghalen
• /etc/fstab: sda in sdb wijzigen en /dev/sda op /space mounten (optional)

If rembrandt1 and 2 have the same kernel, we may want to revert the change of /dev/sda to /dev/sdb.

Monitor

We need to find monitoring program to check if there is no faulty disk

Categories

CategoryTasks