Performance Engineer, Cloudflare

• Deep expertise in the Linux kernel memory-management subsystem: the page allocator, virtual memory and memory mapping, reclaim, and the locking and concurrency that hold it together
• A track record of upstream Linux contribution, ideally including maintainership or sustained work in MM or an adjacent subsystem
• Strong systems programming in C, with the judgement to make high-risk changes to shared infrastructure safely
• Ability to dive deep into a complex codebase while keeping the fleet-wide picture in mind, and to own a problem and its outcome rather than a single patch
• Clear communication with engineers across many teams, since the symptoms of an MM problem usually show up far from the MM code

Desirable

• Delivering this platform with high performance and efficiency depends on advanced Linux kernel and hardware capabilities, and that is where this team works
• The Performance Team owns overall system and application performance across that fleet: resource optimization, instrumentation, and the tooling that lets us see what the hardware is actually doing
• System memory sits alongside CPU cycles, network processing, and storage IOPS as a first-class resource, and how the kernel manages it is increasingly the thing that decides how hot we can safely run every machine we own
• We work upstream first
• Several of us maintain or regularly contribute to the Linux kernel, and our default is to fix problems in mainline rather than carry private patches
• This role is built on that principle
• Using memory efficiently is one of the highest-leverage problems we have
• At our scale, how the kernel's memory-management subsystem behaves under pressure has direct consequences for both performance and cost
• As DRAM prices climb, getting more out of the memory we already own is a real lever on infrastructure spend
• Using custom low-overhead instrumentation on production servers, we have identified several scalability bottlenecks in the kernel memory layer that show up under real workloads
• We are looking for an expert to take on these problems, fix them against production workloads, and drive the changes upstream
• Solving them lifts efficiency and latency across the entire fleet rather than in one place
• Diagnose and fix performance and scalability bottlenecks in the Linux kernel memory-management subsystem, informed by real production telemetry at a scale few environments can offer
• Land your work upstream. We expect MM changes to go to mainline first and want you to keep doing the maintainership and review work that earns standing in the community
• Build and extend low-overhead kernel instrumentation (tracepoints, BPF, perf) so we keep measuring before we change, and prove the impact afterwards
• Partner with kernel, networking, storage, and product engineers across the company to turn observed system behavior into concrete MM improvements
• Set technical direction for how Cloudflare uses, tunes, and evolves kernel memory management as we push utilization higher