5.6 Distributed File Systems

		6.4.1 Booting and Shutting Down

		Perhaps the most inelegant feature of a Beowulf cluster is how you turn it on and turn it off. There is no master switch you can flip to turn on the entire system. Unless you assemble some hardware to power up the system with one flick of the hand (and some people have), you will have to walk by each node and press a button. This isn't so bad for smaller systems, but can become time consuming for larger clusters. Unfortunately, there isn't an inexpensive way to get around this. Shutting down the system is very similar. First you need to shutdown the operating system on each node, and then power down each node. Shutting down the OS on each node isn't too hard. You can do this with the prsh command described in Chapter 5. Just tell it to execute the following shutdown command, or some variation thereof on each node:

		/sbin/shutdown -h now

		If your Beowulf nodes are equipped with ATX power supplies, you can avoid walking over to the machine room and turning each node off by hand. Rather, you can pass the -p option to shutdown, which will power off the nodes for you after the shutdown completes. Rarely will you ever have to completely shutdown and turn off a Beowulf cluster. Once you have turned the machine on, you will likely never turn it off, except for the few nodes that eventually suffer hardware failures or software bugs. It is not uncommon for Beowulfs to go several months without a forced reboot. Some systems are used to run applications that take several weeks to complete, so you should make sure that any such applications either have completed or have been checkpointed before rebooting the system.

		6.4.2 The Node File System

		The Linux operating system follows a convention called the Linux File system Hierarchy Standard.3 Beowulf nodes by default follow the same convention with one or two twists. The main consideration in deciding how to set up node file systems isn't so much how to organize the directory structure, but rather how to partition the disks and what mount points to use for each partition. The primary partitions usually correspond to the following directories:

		3The Linux FHS was formerly known as the Linux File System Standard (FSSTND). The latest version of the Linux FHS as well as the older FSSTND, are published at: http://www.pathname.com/fhs/