Diskless Node - Principles of Operation

Principles of Operation

The operating system (OS) for a diskless node is loaded from a server, using network booting. In some cases, removable storage may be used to initiate the bootstrap process, such as a USB flash drive, or other bootable media such as a floppy disk, CD or DVD. However, the firmware in many modern computers can be configured to locate a server and begin the bootup process automatically, without the need to insert bootable media.

For network auto-booting, the Preboot Execution Environment (PXE) or Bootstrap Protocol (BOOTP) network protocols are commonly used to find a server with files for booting the device. Standard full-size desktop PCs are able to be network-booted in this manner with an add-on network card that includes a UNDI boot ROM. Diskless network booting is commonly a built-in feature of desktop and laptop PCs intended for business use, since it can be used on an otherwise disk-booted standard desktop computer to remotely run diagnostics, to install software, or to apply a disk image to the local hard drive.

After the bootstrapping process has been initiated, as described above, bootstrapping will take place according to one of three main approaches.

• In the first approach (used, for example, by the Linux Terminal Server Project), the kernel is loaded into memory and then the rest of the operating system is accessed via a network filesystem connection to the server. (A small RAM disk may be created to store temporary files locally.) This approach is sometimes called the "NFS root" technique when used with Linux or Unix client operating systems.

• In the second approach, the kernel of the OS is loaded, and part of the system's memory is configured as a large RAM disk, and then the remainder of the OS image is fetched and loaded into the RAM disk. This is the implementation that Microsoft has chosen for its Windows XP embedded remote boot feature.

• In the third approach, disk operations are virtualized and are actually translated into a network protocol. The data that are usually stored in a disk drive are then stored in virtual disks files homed on a server. The disk operations such as requests to read/write disk sectors are translated into corresponding network requests and processed by a service or daemon running on the server side. This is the implementation that is used by Neoware Image Manager, Ardence, VHD and various "boot over iSCSI" products. This third approach differs from the first approach because what is remote is not a file system but actually a disk device (or raw device) and that the client OS is not aware that it is not running off a hard disk. This is why this approach is sometimes named "Virtual Hard Disk" or "Network Virtual Disk".

This third approach makes it easier to use client OS than having a complete disk image in RAM or using a read-only file system. In this approach, the system uses some "write cache" that stores every data that a diskless node has written. This write cache is usually a file, stored on a server (or on the client storage if any). It can also be a portion of the client RAM. This write cache can be persistent or volatile. When volatile, all the data that has been written by a specific client to the virtual disk are dismissed when said client is rebooted, and yet, user data can remain persistent if recorded in user (roaming) profiles or home folders (that are stored on remote servers). The two major commercial products (the one from Hewlett-Packard, and the other one from Citrix Systems) that allow the deployment of Diskless Nodes that can boot Microsoft Windows or Linux client OS use such write caches. The Citrix product cannot use persistent write cache, but VHD and HP product can.