The first thing to remember in Network Booting is that there is more than one computer involved. This has advantages, but it can add complexity. This project is intended to maximize the advantages of network booting, manage the complexity, and address some of the shortcomings of software in the field.
The first goal is to not over simplify. Using Microsoft's tools for network booting works great for booting Windows, can with much work be used to boot Linux, But is completely useless for booting OS-X. Apple's tools are similar in scope if different in target. These and other vendors seek to limit complexity by ignoring it. They say, "This is the way we do it here. We don't care if it does not work with their software, that is outside our scope." Our stance is broader. We want to design a system that works with all the computers in an organization, with an emphasis on open source.
The second goal is user interface. There are three parts to this. First is to have it just work. The user that knows what he wants should be able to do it without difficulty. The second is perceived simplicity. The user should be able in his native language to think no more deeply than he wants to. The third is available complexity. It was better said that "it makes simple things easy and difficult things possible."
Now to limit our scope. The core protocols we are going to use are dhcp and tftp. That is if it does not use those protocols, we will not support it. Additionally we may use http, iscsi, and aoe if they are available. Further we will endorse but not directly use ssh as the remote administration protocol.
To maximize code reuse we will use common open source software wherever possible, contributing changes back where improvements are needed to meet our requirements. Specifically we will use ISC's dhcp server, any tftp server and any http server as our server components. Client side we will use the hardware vendor's boot client whenever possible, iPXE on those machines where it will run but there is not a vendor client, and whatever we can find on other hardware.
Now a brief detour into philosophy. There are three concepts associated with boot loaders, they are boot straps, boot menus, and boot monitors. A boot strap is responsible for loading an operating system. It does not need to present any messages to the user beyond progress messages during normal operation. It may present error messages which are so cryptic that reference works must be consulted to determine their meaning without sacrificing usability, because there is no recovery anyway. A boot menu on the other hand is designed to interact with the user, namely it displays a menu and the user makes a choice. At this point the boot menu either becomes or invokes a boot strap. A boot monitor may also be a boot strap or a boot menu, but is identified by the ability to do specific tasks to configure a system, recover a system that will not boot or do low level diagnostics. Our goal is to use a boot menu wherever possible. A boot strap may be used directly if there is not a open source boot menu available. If there is boot monitor functionality available we will take advantage of it as long as it does not interfere with our other goals. We will make no significant effort to make the boot monitor user friendly, as that falls into the category of hard things possible.
Although we may write boot menus for places where they are not available, We will use existing ones wherever possible, specifically we will use iPXE, grub, and pxelinux. Other boot menus and boot straps are under consideration for other platforms.
The first piece is a configuration file for ISC's dhcp server to allow it to net boot as many different hardware platforms as possible. Also needed are other server side configuration tools to make the preparation of configuration files and boot images as easy as possible.
The second piece is a high level boot language. So far without any effort I have found at least six different languages or notations for writing boot menus. All of them are completely incompatible. I have quit counting, instead I have decided to write a domain specific language for describing basic boot menus that can be compiled to any menu specific language. This is a work in progress.