Linux 5.10 Will Be Able To Hibernate + Resume Much Faster
If you frequently put your system(s) through hibernation cycles, the performance should be much better beginning with the soon-to-start Linux 5.10 cycle.
A change set to land for the upcoming Linux 5.10 merge window allows the batching of hibernate/resume I/O requests in order to speed-up both the hibernation and resume processes.
Up to now the Linux kernel has surprisingly relied on submitting individual I/O requests for each page of data, but now with Linux 5.10+ it's set to use batches with blk_plug.
The patch by Xiaoyi Chen notes, "Testing this change with hibernate and resumes consistently shows merging of the IO requests and more than an order of magnitude improvement in hibernate and resume speed is observed. One hibernate and resume cycle for 16GB RAM out of 32GB in use takes around 21 minutes before the change, and 1 minutes after the change on a system with limited storage IOPS."
Xiaoyi Chen led this work under the Amazon EC2 umbrella but this change is all done in common power management code for system hibernation. Those with lower I/O storage performance will see greater benefits but overall this should be a measurable win come Linux 5.10, which will be kicking off its cycle shortly and likely going to stable around the end of the year.
See this patch part of the power management linux-next material for more details on the batching of I/O requests during hibernate/resume.
A change set to land for the upcoming Linux 5.10 merge window allows the batching of hibernate/resume I/O requests in order to speed-up both the hibernation and resume processes.
Up to now the Linux kernel has surprisingly relied on submitting individual I/O requests for each page of data, but now with Linux 5.10+ it's set to use batches with blk_plug.
The patch by Xiaoyi Chen notes, "Testing this change with hibernate and resumes consistently shows merging of the IO requests and more than an order of magnitude improvement in hibernate and resume speed is observed. One hibernate and resume cycle for 16GB RAM out of 32GB in use takes around 21 minutes before the change, and 1 minutes after the change on a system with limited storage IOPS."
Xiaoyi Chen led this work under the Amazon EC2 umbrella but this change is all done in common power management code for system hibernation. Those with lower I/O storage performance will see greater benefits but overall this should be a measurable win come Linux 5.10, which will be kicking off its cycle shortly and likely going to stable around the end of the year.
See this patch part of the power management linux-next material for more details on the batching of I/O requests during hibernate/resume.
46 Comments