Kernel Developers Discuss Defaulting To BFQ For Some Storage Devices
There's a new discussion taking place over the default I/O scheduler of the Linux kernel.
Since going mainline in Linux 4.12, the BFQ I/O scheduler has continued improving. For those not familiar with the Budget Fair Queueing I/O scheduler, it's designed for low-latency in interactive applications and soft real-time workloads, higher speed and throughput than CFQ/Deadline for many workloads on SSDs, and strong fairness/bandwidth guarantees. BFQ has been used by the default within the Linux kernel "Zen" downstream flavor along in select distributions, but now there's talk again about trying to make it the default I/O scheduler.
Under the proposal led by BFQ maintainer Paolo Valente, BFQ would become the default I/O scheduler for devices having a single hardware queue. This initially would make it the default for MMC/SD cards and other select devices. The renewed discussion is happening in this kernel thread.
Linux block subsystem maintainer Jens Axboe though has expressed a preference to setting this default via udev rules and leaving it more up to Linux distribution vendors to decide their defaults. We'll keep monitoring the discussion and see where it leads.
Since going mainline in Linux 4.12, the BFQ I/O scheduler has continued improving. For those not familiar with the Budget Fair Queueing I/O scheduler, it's designed for low-latency in interactive applications and soft real-time workloads, higher speed and throughput than CFQ/Deadline for many workloads on SSDs, and strong fairness/bandwidth guarantees. BFQ has been used by the default within the Linux kernel "Zen" downstream flavor along in select distributions, but now there's talk again about trying to make it the default I/O scheduler.
Under the proposal led by BFQ maintainer Paolo Valente, BFQ would become the default I/O scheduler for devices having a single hardware queue. This initially would make it the default for MMC/SD cards and other select devices. The renewed discussion is happening in this kernel thread.
Linux block subsystem maintainer Jens Axboe though has expressed a preference to setting this default via udev rules and leaving it more up to Linux distribution vendors to decide their defaults. We'll keep monitoring the discussion and see where it leads.
1 Comment