On 10/9/2011 3:42 PM, vg_ us wrote: > will postmark transaction test do? here - > http://www.phoronix.com/scan.php?page=article&item=linux_2639_fs&num=1 > stop arguing - I think postmark transaction was the only relevant test > XFS was loosing badly - not anymore... > search www.phoronix.com for other tests - there is one for every kernel > version.
This testing was performed on a laptop with a single 7.2k 100GB SATA drive. XFS performance on a single drive is usually irrelevant as that's not its use case. This does however demonstrate the metadata performance improvements. As I've mentioned before, the delayed logging code, in essence, merges individual writes to the log journal with in memory operations, then does a single write out of the total related metadata changes. The on disk result is the same, but the number of physical IOPS required is a fraction of the previous behavior. The one downside to this is potentially more lost metadata in the event of a crash. Considering how much buffering modern operating systems perform this shouldn't come as a shock to anyone. Performance requires buffering for optimized asynchronous operation. This technique is but another example. -- Stan
