Jessie Frazelle, Bryan Cantrill and Steve Tuck have announced the launch of Oxide Computer Company to deliver ‘hyperscaler infrastructure for the rest of us’. The company aims to tackle the ‘infrastructure privilege’ presently enjoyed by hyperscale operators by developing ‘software to manage a full rack from first principles’, including platform firmware.
The founders have extensive cloud experience, with Cantrill and Tuck coming from Joyent, and Frazelle having worked at Docker, Mesosphere, Google and Microsoft. Cantrill describes himself as being ‘haunted by the pain and futility of trying to build a cloud with PC-era systems’. He’s clearly not alone, as Twitter engineers Matt Singer and Nik Johnson previously presented at QCon on their ‘Hardware & Provisioning Engineering’ efforts that created front end and storage servers that were cheaper to own and operate than the off the shelf equipment that they’d previously been using. Initiatives like the Open Compute Project (OCP) have brought open source hardware designs from hyperscale operators like Facebook to the hands of any company willing to engage with original design manufacturers (ODMs); but it appears that those designs and their suppliers haven’t yet displaced the well known original equipment manufacturers (OEMs), that are still churning out servers with designs largely unchanged since the late 90s based on redundant arrays of inexpensive (spinning) disks (RAID).
Frazelle wrote earlier this year in ACM Queue about Open-source Firmware, and the perils lurking in baseboard management controllers (BMCs) and the Unified Extensible Firmware Interface (UEFI) kernel; going on to observe:
Between Ring -2 and Ring -3 there are at least two and a half other kernels in our stack that have many capabilities. Each of these kernels has its own networking stacks and web servers, which is unnecessary and potentially dangerous, especially if you do not want these rings reaching out over the network to update themselves. The code can also modify itself and persist across power cycles and reinstalls. There is very little visibility into what the code in these rings is actually doing, which is horrifying, considering these rings have the most privileges.
In his ‘The soul of a new computer company’ post Cantrill references his QCon presentation ‘Is It Time to Rewrite the Operating System in Rust?’ in noting that the name Oxide in part references their expectation that Rust will be ‘playing a substantial role for us’. So Oxide intends to firstly simplify the firmware environment by eliminating elements that Joshua M. Clulow describes as ‘complicated to implement and to consume’, whilst at the same time taking advantage of the better memory safety of Rust (over C) to avoid creating vulnerabilities.
Oxide’s launch posts don’t just talk about their technology plans, but also cover their Principles of Operation that enumerate the principles and values that the company will operate by.
Alongside the company launch the founders have also debuted their On the Metal podcast featuring interviews with Facebook’s first VP of Engineering Jeff Rothschild and Salesforce VP of Infrastructure Engineering Amir Michael. The interviews touch upon some of Oxide’s plans along with tales of hacking up and down the hardware and software stack.