Linux Writecache To See Much Greater Performance On Intel Optane Systems Soon
The Linux DM-Writecache target that allows for writeback caching to cache newly written data to an SSD or persistent memory will with Linux 5.8 see better performance out of Intel Optane like storage on newer platforms.
Red Hat developer Mikulas Patocka has been working to enhance the dm-writecache performance on DDR-based persistent memory, namely Intel Optane products.
The performance optimization now queued for Linux 5.8 is making use of CLFLUSHOPT within dm-writecache when available. CLFLUSHOPT is one of Intel's persistent memory extensions that allows for optimized flushing of cache lines by supporting greater concurrency. The CLFLUSHOPT instruction has been supported on Intel servers since Skylake and on AMD since Zen.
The dm-writecache target will now check for CLFLUSHOPT support and use it when available, thereby helping the performance on Optane-like storage for this writeback caching. On unsupported CPUs, the existing behavior is maintained.
With a 1k block size, making use of CLFLUSHOPT in dm-writecache over MOVNTI boosted the performance from 642 to 688 MB/s, for 2k block sizes from 725MB/s to 1.1GB/s and for 4K block sizes from 744MB/s to 1.2GB/s.
The patch is queued as part of the Device Mapper changes coming up for the Linux 5.8 cycle that will be kicking off next month.
Red Hat developer Mikulas Patocka has been working to enhance the dm-writecache performance on DDR-based persistent memory, namely Intel Optane products.
The performance optimization now queued for Linux 5.8 is making use of CLFLUSHOPT within dm-writecache when available. CLFLUSHOPT is one of Intel's persistent memory extensions that allows for optimized flushing of cache lines by supporting greater concurrency. The CLFLUSHOPT instruction has been supported on Intel servers since Skylake and on AMD since Zen.
The dm-writecache target will now check for CLFLUSHOPT support and use it when available, thereby helping the performance on Optane-like storage for this writeback caching. On unsupported CPUs, the existing behavior is maintained.
With a 1k block size, making use of CLFLUSHOPT in dm-writecache over MOVNTI boosted the performance from 642 to 688 MB/s, for 2k block sizes from 725MB/s to 1.1GB/s and for 4K block sizes from 744MB/s to 1.2GB/s.
The patch is queued as part of the Device Mapper changes coming up for the Linux 5.8 cycle that will be kicking off next month.
5 Comments