Managing RAID on Linux - Derek Vadala [34]
Figure 2-29. My web server contains a quad-channel SCSI controller. Three disks are connected to each channel.
Hardware is always changing and the equipment you buy doesn't always meet your expectations, so it's always a good idea to do research before building or purchasing any system.
As an example of what can go wrong, a former colleague recently told me that he had to argue with his vendor in order to get a system with multiple SCSI backplanes. He had ordered a dual-channel RAID controller in a rackmount case with eight hard disks. But the vendor had designed the system so that there was only a single SCA backplane. This meant that all of his data would be travelling over a single SCSI channel and that the second channel would be wasted. The vendor offered the option of adding an external disk-only case for the second channel, but my colleague found that unfeasible due to the the high price of server colocation it added. In the end, my colleague had to swap his components into a new case with two backplanes. The vendor ate the cost, but it took an extra week to get the system online.
Also, remember compatibility issues. I recommend checking relevant mailing lists and web sites to make certain that your disk controllers will work properly with your motherboard and network controller.
Chapter 3. Getting Started: Building a Software RAID
Before plunging into this chapter, you should be familiar with the process of building and installing a new kernel. If you have never done this before, I strongly suggest that you read Brian Ward's "The Linux Kernel HOWTO" (http://www.tldp.org/HOWTO/Kernel-HOWTO.html), which is available at any Linux Documentation Project mirror and is likely included with your distribution. If you prefer books to online documentation, then I'd recommend you pick up a copy of Running Linux (O'Reilly). Chapter 7 of that book offers an excellent tutorial on managing the kernel. To become comfortable installing a new kernel, I suggest you make some simple changes first.
A good start would be to eliminate some loadable modules from your kernel and include those subsystems statically. Most distributions set up major system peripherals as loadable modules, rather than compiling them statically into the kernel. Network drivers are a good example of kernel subsystems that are often installed, by default, as loadable modules. The sheer number of available network hardware configurations makes this the only efficient way to circulate network-enabled Linux distributions. So, in order to maintain compatibility with as many systems as possible, distributors such as Red Hat, Mandrake, and SuSE automatically load modules appropriate to your system at boot time, while installing a stock kernel with support for loadable modules. Using modules also helps conserve system memory by reducing the size of the running kernel. Modules can be unloaded when they are no longer needed, freeing up additional system resources.
While this is a completely viable way to set up a Linux machine, I tend to feel that a monolithic kernel, a kernel without modules, is more appropriate for system stability and is essential for servers. Loadable modules leave room for uncertainty during system initialization and are therefore best left to desktop machines and novices. Since servers typically perform specific and dedicated services, a monolithic kernel suits them best. Once you are comfortable recompiling the kernel and reconfiguring your boot manager, you can begin adding support for RAID.
This chapter will guide you through implementing a software RAID by explaining how to:
Enable kernel RAID features
Download and install software RAID utilities
Partition