One of the most touted benefits of Linux and open source programs is their flexibility. However, as the popularity of Linux has grown, some of the flexibility seems to have been sacrificed. As larger Linux vendors have become more standardized to support certified applications, the freedom to mold your Linux distribution to your needs has diminished a bit.

Adding an unsupported kernel module or otherwise modifying your distro may void your support contract or introduce other problems that are not easily resolved. Even larger, well-established independent software vendors who wish to add Linux as a supported platform might be not be able to preserve their customers' Linux support contracts if they require specialization that exceeds the confines of the supported distribution, even if there are no technical limitations to doing so.

Billy Marshall, co-founder of rPath, started to see this trend as the vice president of North American sales for Red Hat. Software vendors such as IBM were running into problems in which the stock versions of Red Hat Linux needed enhancements to optimize performance for certain packages, but had the potential to compromise other systems. Before joining rPath, Marshall says he started to witness three market trends that validated his belief in the need for rPath products. His account of these trends involved three main elements.

Linux and open source continued to be popular and more open source applications were becoming interesting. The second trend, Software as a Service, was gaining popularity. People really didn't care what infrastructure they ran as long as they got the value of the application. Users were willing to give up control of the Lincoln Logs or Legos that were making up the operating system as long as it worked the way they wanted. Salesforce.com was hitting their stride but nobody was asking them what OS or version they where using. They only cared whether the application worked. The last trend we noticed was virtualization. VMware was experiencing extraordinary growth, and we wanted to address the need for virtualized applications. Once the hardware layer is abstracted, you can now put multiple applications on the same box without interrupting each other or having to run the same operating system.

With these observations in hand, Marshall wanted to solve the problem of trying to be all things to all people, so he partnered with former Red Hat colleague Erik Troan to develop a technology that automated the development of made-to-order Linux distributions. They developed task-based Linux distributions that where targeted at specific needs of individual users (usually ISVs or wannabe appliance vendors) rather than a broad platform that encompassed 80% the needs of many, many users.

Troan and Marshall promote rPath as a software appliance company that focuses on developing task-based Linux distributions specifically tailored to users' needs. The rPath approach puts a new twist on Linux build systems by trying to build Linux distributions that most closely match the needs of the end user rather than trying to build a system that a great number of users could take advantage of.

Their model is to provide Software as a Service by helping companies develop and maintain repositories for their custom applications. This could be a distribution packaged with a specific application as they do for open source PBX vendor, Digium. Or it could be virtual machine images that can be downloaded and run on VMware or popular community virtualization software QEMU and, eventually, the open source Xen virtualization solution. Their ability to provision and maintain task-based operating systems augmented with their online interface makes rPath unique.

Conary - A New Package Management System
Most Linux administrators are used to updating their Linux distribution through the RPM or apt system. Using these systems they often suffer from mismatched software maintenance streams, a bane of system administrators. Keeping systems up to date also introduces the perils of system conflicts and bloated packages that include not only the essential libraries, bit, and bytes, but the unnecessary extras. Conary is more utilitarian in its approach. It is a distributed software management system that allows a group of loosely connected repositories to define components to be installed into the Linux system. These components can be more granular than widely distributed RPM or dpkgs (dee-packages).

Ironically the man who authored the popular RPM package management system, rPath co-founder Erik Troan, was the one who developed this new software provisioning system. According to Troan: "Open source is supposed to be about flexibility. So our base technology takes entire open source projects, puts them together into useful systems that are flexible, containable, and updatable." Conary works by using a versioned repository or set of repositories that host source codes and binary files, whereas traditional RPM and apt repositories are collections of pre-packaged files. In turn, components are then combined into packages that giving users more granular control of what is included in their packages.

Obviously developers could make their own systems packages, but that's not where their time is best spent. They also would have to track versions they build. Where RPMs are usually identified by name and version, Conary uses systematic versions to avoid confusion in all aspects of the system. Since the packages are collections of files in a repository, the version is specified as the repository location, then the original version number (from the authors of the software), the source revision number, and then the binary build revision number. The power in this is that it allows for branch distribution where developers change only pieces of their distribution. Development streams can now not only diverge but converge at the places appropriate for the application. ISVs can then focus on adding value to their applications without having to duplicate the non-application-specific parts of their product.

rPath in Action: rBuilder Online
rBuilder Online (www.rpath.com/rbuilder/) is the free-to-use tool that you can use to build your own Linux distribution and include open source software of your choosing. Most projects start with rPath, the reference architecture for rPath-built Linux distros. You simply search the available projects and packages to combine the elements to make your own distribution. You can register as a product maintainer complete with mailing lists and project page to enlist others to help maintain your project. The three simple steps to rBuilder Online are to find what you need, build your recipe, and finally cook an ISO or other type of image and download it (see Figure 1).

It really is that simple. It's also a great way to demonstrate how your build process could work if you were an rPath customer. Which leads to the obvious question in most open source businesses, how do these guys make money? Simple. For commercial vendors, rPath will offer support and other services for application vendors. This could include offering your own update repository and mechanism for your customers complete with rPath testing and quality assurance for the packages they supply.

Summary
rPath has settled on the term the software appliance to try to convey their vision, though what they offer is a way to provision task-based Linux distributions through a clever Linux-build technology. Today they provide operating systems and maintenance for custom-built Linux distributions that will likely be used in embedded devices or appliances. Though their technology has the ability to transcend these boundaries, a recent deal with EMC-owned VMware announced rPath's freely available Virtual Appliances that included Apache, LAMP, SugarCRM, and Port25 appliances that could be run from VMware virtualization products offered through the VMware Technology Network (www.vmware.com/vmtn/).

Ultimately, you could provision applications built-to-order using rPath tools that run only the application and what is required to support them. This has some very attractive benefits with Linux servers, the first being installation. End users receive built-to-order images that install the operating system and applications in one fell swoop. The second benefit is conflict resolution: by allowing the intelligence to reside in rPath's build process installation, there should be no need to track down supporting packages to allow your application to function optimally. Finally, maintenance: you get all the operating system and supporting open source packages from one single source, rPath, a good alternative to traditional operating system providers.