ALS: Automotive Grade Linux

Using Linux in cars is a hot topic, even if the market is less visible to most developers than tablets or mobile phones. The Linux Foundation (LF) announced an initiative at the second Automotive Linux Summit in Gaydon, UK, however, that may result in a higher profile for automotive Linux development. The initiative is called Automotive Grade Linux (AGL), and its goal is to produce a distribution tuned for deployment throughout a vehicle, including in-dash systems, instrument clusters, and even safety-critical engine control units. A number of automakers and industry players are on board — which sparked some confusion at the announcement, because many of the same companies are also involved with existing Linux-based automotive efforts like GENIVI.

AGL announced

LF Executive Director Jim Zemlin announced AGL in an Automotive Linux Summit keynote on September 19. Three automakers are founding participants: Toyota, Nissan, and Jaguar Land Rover. They are joined by a number of electronics and silicon vendors, including Texas Instruments, Intel, Samsung, and Fujitsu. Officially, AGL is a "workgroup," as distinguished from a software project. Zemlin likened it to Carrier Grade Linux, a workgroup started by telecommunications companies in 2002 to address that industry's needs as it migrated its equipment to Linux from proprietary operating systems.

The AGL announcement states that the workgroup "will facilitate widespread industry collaboration that advances automotive device development, providing a community reference platform that companies can use for creating products." That reference platform, it continues, will be a Tizen-based distribution "optimized for a broad set of automotive applications ranging from Instrumentation Cluster to In-Vehicle-Infotainment (IVI) and more." The announcement specifically mentions fast boot and extended lifecycle support for automotive products as features, and says that the workgroup will support other industry efforts like GENIVI and the W3C's Web and Automotive workshop.

During the Summit, a number of people — speakers included — expressed puzzlement about AGL, specifically with regard to what its ultimate "deliverables" will be, and to how exactly it competes or cooperates with the other automotive Linux efforts like GENIVI and Tizen's IVI platform. Zemlin noted in his keynote that there is no automotive-focused equivalent to the community-based distributions like Debian and Fedora, and said that as a result it is much more difficult for interested community developers to get started working on the automotive-specific problems faced by carmakers and product vendors. There is now an AGL site alive at automotive.linuxfoundation.org, which provides a bit more detail, and references that same issue on its "About" page. It compares the community-managed Debian and Fedora to the commercially-supported Ubuntu and Red Hat Enterprise Linux, and says "In a similar manner, AGL is seeking to become the upstream Linux distribution for automotive use by facilitating cooperation between multiple industries and the open-source communities."

So, then, the "product" to be produced by AGL would appear to be a full-fledged Linux distribution, rather than a suite of platform packages or a specification. As to the scope of the project, the site also says AGL is not limited to IVI systems, but also encompasses "instrument clusters, climate control, intelligent roadway instrumentation, etc." The site also sets out a project structure, including a steering committee, steering committee coordinator, and various expert groups tasked with developing specific features. The makeup of the committee and the specifics of the expert groups have not been announced; there are, however, two public mailing lists available (in addition to a private one for the steering committee).

Whither GENIVI?

Although the announcement and site both say that AGL is not a challenger to GENIVI, it is not difficult to see why some people (particularly those working on GENIVI) either perceive the projects as potential competitors or fear a duplication of effort. Both, after all, are automotive industry associations attempting to build a Linux-based platform that meets the shared requirements of car manufacturers and tier-one equipment makers (and indeed quite a few industry players are members of both efforts). Both target Linux and core services that need to be adapted from the desktop/server/mobile markets where Linux is already established, and both envision their software as some sort of "reference implementation." GENIVI's output is a baseline which is "respun" into other distributions, while AGL's is an "upstream" distribution intended to be adapted and optimized in products.

Still, as similar as that language sounds, there are some arguably important details that distinguish the two projects' goals. First, GENIVI is ultimately a compliance-driven specification: the baseline software that it creates en route is simply a means to that end. This process can be confusing, in large part because both the specification itself and the compliance process are closed to non-GENIVI members. Consequently, those on the outside primarily see the commercial products and distributions that reach compliance.

Second, GENIVI is targeting a middleware platform only. That is to say, the purpose of certifying a particular software stack as GENIVI compliant is that it offers guarantees regarding application- and service-level compatibility. As Visteon's Pavel Konopelko explained in his session, the specification includes numerous "abstract" and "placeholder" components. For example, the Bluetooth stack could be Linux's native BlueZ or a proprietary replacement; either would qualify as long as it implements the required functionality. In addition, GENIVI has not tackled lower-level topics like Controller Area Network (CAN) bus support. CAN bus is a common transport mechanism, but it sits well below the application layer.

Of course, CAN bus may be on its way out; the protocol offers no security and certainly lacks the flexibility of standard TCP/IP. But because GENIVI is also focused on IVI systems specifically, inter-device communication is a bit of a tangent. A third difference between the projects is that AGL draws a wider circle, encompassing non-IVI components. Over the course of the Summit, there were talks about other automotive computing issues, such as communicating with intelligent roadways — e.g., to automatically relay speed limit information or safety reports. Jaguar Land Rover operated an exhibit at the summit's venue, the British Motor Heritage Center, that focused on its new vehicles' automatic adjustments to suspension, braking, and handling in response to off-road conditions. Such things are certainly outside the purview of IVI and, like engine control units (ECUs), probably even more meticulously scrutinized by company lawyers.

The other side to the answer is that AGL bills itself as an open collaboration project, while GENIVI is still members-only. There appears to be movement toward additional openness from GENIVI, and several GENIVI speakers alluded to forthcoming progress on that front at the summit. Of course, AGL has yet to get rolling; it is always possible that the corporate membership will be more secretive than the volunteer free software contributor community would like as well.

Tizen, workgroups, and collaboration

Another factor worth assessing is how AGL will affect the Tizen project. Tizen's two main supporters, Intel and Samsung, are AGL members as well, and the AGL project has already announced that it will use Tizen as the basis of its distribution. On the one hand, this seems to make AGL both an "upstream distribution" to its corporate adopters and a "downstream distribution" to the Tizen Project, which otherwise appears unchanged. On the other hand, perhaps seeing Tizen used as the basis of AGL's distribution work will make Tizen's insistence that it is a "platform" and not a distribution itself a little easier to parse.

Then again, what constitutes a platform and what constitutes a distinct distribution is largely a word game (for proof of that, consider the ever-expanding litany of X-as-a-Service acronyms generated by the cloud computing sector). Tizen remains committed to offering a Linux system that consumer device makers can build on in multiple categories. Tizen (and MeeGo before) it have been advertising such flexible functionality for two years or so, but the automotive market has always seemed to be the ripest for adoption. We may not see Tizen-based phones in the near future, and TVs or set-top boxes are likely to not sport platform branding at all, so perhaps focusing on automotive Linux is the quickest path to success anyway. The difficulty will be managing AGL's insistence that it is building a distribution for IVI and non-IVI automotive computing. The Tizen and MeeGo efforts were explicitly IVI-focused, and skeptics could be forgiven for wondering if Tizen's HTML5 application platform is sufficient for safety-critical uses like dashboard instrument clusters.

One attendee at the summit joked privately that AGL was probably formed because Toyota wanted to be in the driver's seat (pardon the expression). That is a bit cynical if taken at face value, but even if it were true, the LF does exist to accommodate companies that are new to collaborating around Linux. Periodically that may mean hosting a workgroup (such as Carrier Grade Linux or the Consumer Electronics workgroup (CELF)) that seems quite a ways outside the mainstream community. What matters in the long run, however, is that most of these companies eventually become mainstream contributors to the kernel and other parts of the standard Linux stack. Those companies may have unease about working with free software, or about collaborating with their competitors, but often these industry efforts produce work that benefits the rest of the community. The Long Term Support Initiative, for example, grew out of CELF.

It was clear from the Automotive Linux Summit that the car industry is ready to migrate to Linux as quickly as it can manage the transition; the costs of developing and supporting proprietary systems add up more quickly in automotive than they do in most other fields, in no small part because of the decade-long lifecycle of the automobile. Car-buyers expect their vehicles to be serviceable (and, in fact, dealer-serviceable) for ten or more years, a situation that Matt Jones of Jaguar Land Rover said led to his company's current burden of simultaneously supporting three unrelated IVI platforms at different times in recent years. At the moment, the launch of AGL may seem to crowd in on GENIVI, but there is no shortage of development to be done. Besides, who knows? Three or four years from now the two projects may have enough in common to work hand-in-hand or to merge — yet that will still be less than halfway through the lifespan of a typical automotive computer.

[The author would like to thank the Linux Foundation for travel assistance to ALS.]