I work between product direction and the systems that have to deliver it.

A product manager is an interface between layers that do not always describe the same problem in the same way.

I set strategic direction for products based around storage and GPUs. The work sits between customer and commercial needs, product priorities, engineering choices and the behaviour of the system itself. My job is to make the decision clear without hiding the constraint that could overturn it.

That is why I stay close to the technical detail. A capacity figure, benchmark or compatibility claim is only a starting point. I want to know how it was measured, which workload produced it, what happens during recovery and whether the result survives a longer run.

The storage material covers NVMe, NVMe-oF, SPDK, RDMA, RoCE and S3 testing. The same way of thinking appears in my Ceph Days London presentation: begin with the workload and operational consequence, then choose the hardware.

Strix Halo is the current investigation. Its software is changing quickly as support expands, which makes compatibility, performance, correctness, memory behaviour and stability separate questions. I publish the machine, versions, commands, datasets and failures so that readers can see how I reached the answer and where it may no longer apply.

The subjects on this site will change. The method will not: question the claim, test what can be tested, explain the mechanism and be clear about what remains unknown.