Open Source Summit + Embedded Linux Conference North America 2026

Flash memory is the literal foundation of an embedded system, yet it is a finite resource. Every log entry, database commit, and firmware update inches the device closer to its end of life. For developers managing fleets of devices, the question is not just if the flash will fail, but when and which process is the culprit.

This session dives deep into the lifecycle of a write, from a high level look at the physics behind flash memory, to how we can get an estimation of lifetime by tracking number of bytes written. We will start at the hardware level, explaining the physical degradation of NAND cells and why eMMC controllers use wear leveling to manage this reality. Next, we will bridge the gap between hardware specs and software reality using the Total Bytes Written (TBW) metric to estimate remaining life.

Moving into the Linux kernel, we will explore the built-in metrics found in procfs and sysfs to monitor disk I/O. Finally, we will level up our observability by using eBPF to build a per process "write shaming" tool. This allows us to pinpoint exactly which application or daemon is burning through our hardware lifespan.