Advanced Computing in the Age of AI | Tuesday, March 19, 2024

Red Hat Rolls Up Linux For ARM Servers 

The great thing about Linux is that it runs on nearly everything, and if ARM servers ever take off, they will do so because Linux workloads are ported from X86 and other architectures to run on 64-bit ARM processors. The classic chicken and egg problem is making it difficult for an ARM server ecosystem to develop, and Red Hat is going to help this along with an effort it calls the ARM Partner Early Access Program.

That is a boring name for an effort that will give ARM chip makers and their server partners a variant of the Red Hat Linux software stack that is tuned up to run on the latest 64-bit ARM processors. This variant of Red Hat Linux does not have a name yet, says Mark Coggin, senior director of product marketing for the Platform Business Unit at Red Hat. The partner program will give companies access to something that is conceptually somewhere between the Fedora development release, which has supported ARM processors for several iterations, and the commercial-grade Red Hat Enterprise Linux, which does not yet support ARM, even with the just-announced RHEL 7 version. RHEL 5 supported X86 platforms as well as Itanium, Power, and IBM System z mainframe platforms, and RHEL 6 dropped Itanium support. To be supported by RHEL, a platform has to adhere to standards sufficiently that Red Hat can support servers using a particular processor without having to hand-code kernels and drivers for it. This is why Red Hat has been particularly insistent in pushing ACPI, UEFI and Secure Boot standards across the ARM server ecosystem and in developing an ARM variant of Fedora in conjunction with the Linaro Enterprise Group. Red Hat is also one of the key drivers behind the Server Base System Architecture that was announced last year to create a baseline hardware compatibility across 64-bit ARM processors and their servers.

At the moment, only Applied Micro's X-Gene1 and AMD's "Seattle" Opteron A1100 chips are shipping with 64-bit support and aimed at servers. X-Gene1 is available on development boards but has yet to ship in volume, and AMD doesn't expect to get the Seattle chip out the door in volume until the fourth quarter of this year. Networking chip maker Cavium is working on its "Thunder" 64-bit ARM server chip, and rival Broadcom, which has the lion's share of Ethernet switch ASICs these days, is working on its own 64-bit ARM chip, code-named "Vulcan." Nvidia and Samsung Electronics have apparently backed away from their 64-bit server chip efforts, and of course, ARM server chip upstart Calxeda went bust late last year before it transitioned from 32-bit to 64-bit with its EnergyCore processors.

Suffice it to say, the ecosystem is still evolving, and no one is sure who will be the first adopters of the ARM server chips. The odds favor hyperscale datacenter operators, who control their own software stacks from top to bottom, who tend to make large orders for servers, who design their own systems and farm out the manufacturing, and who are always looking for an edge. These companies deploy on Linux almost exclusively (with the exception of Microsoft, of course), and they may benefit from the early partner program that Red Hat is setting up, even if only tangentially instead of directly.

"Most of what we are doing here is seeing where the market is going and where the partners are going to take it," Coggin tells EnterpriseTech. It is hard to predict when the ARM server ecosystem will start delivering iron that customers can deploy in production, and Coggin wants to make it absolutely clear that the PEAP is not a promise to deliver a variant of RHEL for the ARM architecture. "This is not a product commitment. It is an unsupported release of software for partners and we are purposely avoiding a name for it now."

Nonetheless, Red Hat has reasons for not just pushing everything relating to ARM servers through the Fedora community. For one thing, the pace of change with ARM server chips is very fast right now, and it is very likely that this ARM variant will be iterated on a much more frequent basis compared to Fedora. This ARM release will standalone, at least for now, also because the ten-year support guarantee for a RHEL version might not make sense given the pace of change. Equally importantly, the ARM variant could end up using a different version of the Linux kernel and having a different set of add-on programs from the thousands in the RHEL stack. Red Hat does not control how the projects who create these add-ons for X86 and other architectures will port their code to ARM and tune it. Red Hat will, of course, deliver a full-on RHEL release once the ARM server foundation is established and there are significant enough volumes of ARM servers being acquired to make it economically feasible.

Presumably any tweaks that are made with this unnamed Red Hat Linux release will be pushed upstream to the Fedora community on a regular basis and make their way into other Linux distributions. (Canonical is very aggressively supporting ARM processors with Ubuntu Server, and SUSE Linux has also woven ARM support into its openSUSE development release.)

All of the key players in the evolving ARM server ecosystem have signed up to be a member of the ARM Partner Early Access Program, including chip makers AMD, Applied Micro, Broadcom, and Cavium, as well as ARM Holdings, which is the company that sets the ARM chip standards and licenses the core technology to these and other ARM chip makers. BIOS maker American Megatrends has also joined up, as have dominant server makers Dell and Hewlett-Packard, who have been dabbling in ARM servers for years, and the Linaro Linux development organization.

EnterpriseAI