Optimizing Linux MD Bitmap Code Yields 89% Throughput Boost For Quad SSDs
([Linux Storage] 5 Hours Ago
+89.4%)
- Reference: 0001511457
- News link: https://www.phoronix.com/news/Optimize-Linux-MD-Bitmap-RFC
- Source link:
A promising patch for the Linux kernel is optimizing the locking contention and scattered address space for the MD bitmap code to improve both the storage throughput and latency.
The results of the "request for comments" batch to enhance the md/bitmap code is mighty impressive. Here's straight from the patch message:
"Optimize scattered address space. Achieves significant improvements in both throughput and latency.
Maximize thread-level parallelism and reduce CPU suspension time caused by lock contention. Achieve increased overall storage throughput by 89.4% on a system with four PCIe 4.0 SSDs. (Set the iodepth to 32 and employ libaio. Configure the I/O size as 4 KB with sequential write and 16 threads. In RAID5 consist of 2+1 1TB Samsung 980Pro SSDs, throughput went from 5218MB/s to 9884MB/s.)"
This ~200 line optimization patch comes from a Chinese research paper over optimizing Linux software RAID for next-generation storage devices. With the work from the Chinese university researchers they improve throughput by nearly 90% while also decreasing the 99.99th percentile latency by 85% compared to the existing MD RAID code.
Those interested can find the patch and more details via [1]the Linux kernel mailing list while this MD optimization patch undergoes review.
[1] https://lore.kernel.org/lkml/DM6PR12MB319444916C454CDBA6FCD358D83D2@DM6PR12MB3194.namprd12.prod.outlook.com/
The results of the "request for comments" batch to enhance the md/bitmap code is mighty impressive. Here's straight from the patch message:
"Optimize scattered address space. Achieves significant improvements in both throughput and latency.
Maximize thread-level parallelism and reduce CPU suspension time caused by lock contention. Achieve increased overall storage throughput by 89.4% on a system with four PCIe 4.0 SSDs. (Set the iodepth to 32 and employ libaio. Configure the I/O size as 4 KB with sequential write and 16 threads. In RAID5 consist of 2+1 1TB Samsung 980Pro SSDs, throughput went from 5218MB/s to 9884MB/s.)"
This ~200 line optimization patch comes from a Chinese research paper over optimizing Linux software RAID for next-generation storage devices. With the work from the Chinese university researchers they improve throughput by nearly 90% while also decreasing the 99.99th percentile latency by 85% compared to the existing MD RAID code.
Those interested can find the patch and more details via [1]the Linux kernel mailing list while this MD optimization patch undergoes review.
[1] https://lore.kernel.org/lkml/DM6PR12MB319444916C454CDBA6FCD358D83D2@DM6PR12MB3194.namprd12.prod.outlook.com/
bezirg