Sent out today as a request for comments is a new patch series for Dynamic Housekeeping and Enhanced Isolation (DHEI). DHEI aims to provide run-time adjustments to kernel behavior around CPU isolation for helping with latency-sensitive tasks. The expressed goal is for helping cloud-native orchestrators and high frequency trading platforms dynamically re-partition CPU resources without downtime.
Qiliang Yuan of China Telecom sent out the patches today for Dynamic Housekeeping & Enhanced Isolation. Ultimately it’s about features like isolcpus and nohz_full of the kernel being only configurable at boot time via parameters. Thus making changes require a reboot and the associated downtime. With DHEI, the hope is to be able to dynamically re-partition CPU resources for running Linux servers without that reboot/downtime.
The patches sent out today are under a request for comments “RFC” flag with no upstream Linux stakeholders yet commenting on the approach. With the proposed DHEI code, the kernel’s housekeeping boundaries can be manipulated at runtime via /sys/kernel/housekeeping/.
“Key Features:
– Fine-grained control: Separate sysfs nodes for timer, rcu, tick, workqueue, kthread, etc.
– Dynamic NOHZ_FULL: Supports enabling/disabling full dynticks mode on-the-fly by re-kicking CPUs to evaluate tick dependencies.
– SMT Awareness: An optional ‘smt_aware_mode’ ensures that all SMT siblings of a physical core stay in the same isolation state.
– Safety Guard: Prevents the isolation of all CPUs, ensuring at least one online CPU is always available for housekeeping tasks.This series provides the necessary infrastructure for cloud-native orchestrators and high-frequency trading platforms to dynamically re-partition CPU resources without incurring the downtime of a reboot.”
Those wishing to learn more about this DHEI proposal can do so via the RFC patch series.
