linux/fs/btrfs/extent-tree.h

169 lines
5.7 KiB

btrfs: introduce size class to block group allocator The aim of this patch is to reduce the fragmentation of block groups under certain unhappy workloads. It is particularly effective when the size of extents correlates with their lifetime, which is something we have observed causing fragmentation in the fleet at Meta. This patch categorizes extents into size classes: - x < 128KiB: "small" - 128KiB < x < 8MiB: "medium" - x > 8MiB: "large" and as much as possible reduces allocations of extents into block groups that don't match the size class. This takes advantage of any (possible) correlation between size and lifetime and also leaves behind predictable re-usable gaps when extents are freed; small writes don't gum up bigger holes. Size classes are implemented in the following way: - Mark each new block group with a size class of the first allocation that goes into it. - Add two new passes to ffe: "unset size class" and "wrong size class". First, try only matching block groups, then try unset ones, then allow allocation of new ones, and finally allow mismatched block groups. - Filtering is done just by skipping inappropriate ones, there is no special size class indexing. Other solutions I considered were: - A best fit allocator with an rb-tree. This worked well, as small writes didn't leak big holes from large freed extents, but led to regressions in ffe and write performance due to lock contention on the rb-tree with every allocation possibly updating it in parallel. Perhaps something clever could be done to do the updates in the background while being "right enough". - A fixed size "working set". This prevents freeing an extent drastically changing where writes currently land, and seems like a good option too. Doesn't take advantage of size in any way. - The same size class idea, but implemented with xarray marks. This turned out to be slower than looping the linked list and skipping wrong block groups, and is also less flexible since we must have only 3 size classes (max #marks). With the current approach we can have as many as we like. Performance testing was done via: https://github.com/josefbacik/fsperf Of particular relevance are the new fragmentation specific tests. A brief summary of the testing results: - Neutral results on existing tests. There are some minor regressions and improvements here and there, but nothing that truly stands out as notable. - Improvement on new tests where size class and extent lifetime are correlated. Fragmentation in these cases is completely eliminated and write performance is generally a little better. There is also significant improvement where extent sizes are just a bit larger than the size class boundaries. - Regression on one new tests: where the allocations are sized intentionally a hair under the borders of the size classes. Results are neutral on the test that intentionally attacks this new scheme by mixing extent size and lifetime. The full dump of the performance results can be found here: https://bur.io/fsperf/size-class-2022-11-15.txt (there are ANSI escape codes, so best to curl and view in terminal) Here is a snippet from the full results for a new test which mixes buffered writes appending to a long lived set of files and large short lived fallocates: bufferedappendvsfallocate results metric baseline current stdev diff ====================================================================================== avg_commit_ms 31.13 29.20 2.67 -6.22% bg_count 14 15.60 0 11.43% commits 11.10 12.20 0.32 9.91% elapsed 27.30 26.40 2.98 -3.30% end_state_mount_ns 11122551.90 10635118.90 851143.04 -4.38% end_state_umount_ns 1.36e+09 1.35e+09 12248056.65 -1.07% find_free_extent_calls 116244.30 114354.30 964.56 -1.63% find_free_extent_ns_max 599507.20 1047168.20 103337.08 74.67% find_free_extent_ns_mean 3607.19 3672.11 101.20 1.80% find_free_extent_ns_min 500 512 6.67 2.40% find_free_extent_ns_p50 2848 2876 37.65 0.98% find_free_extent_ns_p95 4916 5000 75.45 1.71% find_free_extent_ns_p99 20734.49 20920.48 1670.93 0.90% frag_pct_max 61.67 0 8.05 -100.00% frag_pct_mean 43.59 0 6.10 -100.00% frag_pct_min 25.91 0 16.60 -100.00% frag_pct_p50 42.53 0 7.25 -100.00% frag_pct_p95 61.67 0 8.05 -100.00% frag_pct_p99 61.67 0 8.05 -100.00% fragmented_bg_count 6.10 0 1.45 -100.00% max_commit_ms 49.80 46 5.37 -7.63% sys_cpu 2.59 2.62 0.29 1.39% write_bw_bytes 1.62e+08 1.68e+08 17975843.50 3.23% write_clat_ns_mean 57426.39 54475.95 2292.72 -5.14% write_clat_ns_p50 46950.40 42905.60 2101.35 -8.62% write_clat_ns_p99 148070.40 143769.60 2115.17 -2.90% write_io_kbytes 4194304 4194304 0 0.00% write_iops 2476.15 2556.10 274.29 3.23% write_lat_ns_max 2101667.60 2251129.50 370556.59 7.11% write_lat_ns_mean 59374.91 55682.00 2523.09 -6.22% write_lat_ns_min 17353.10 16250 1646.08 -6.36% There are some mixed improvements/regressions in most metrics along with an elimination of fragmentation in this workload. On the balance, the drastic 1->0 improvement in the happy cases seems worth the mix of regressions and improvements we do observe. Some considerations for future work: - Experimenting with more size classes - More hinting/search ordering work to approximate a best-fit allocator Signed-off-by: Boris Burkov <boris@bur.io> Signed-off-by: David Sterba <dsterba@suse.com>
2022-12-15 16:06:33 -08:00
btrfs: allow to run delayed refs by bytes to be released instead of count When running delayed references, through btrfs_run_delayed_refs(), we can specify how many to run, run all existing delayed references and keep running delayed references while we can find any. This is controlled with the value of the 'count' argument, where a value of 0 means to run all delayed references that exist by the time btrfs_run_delayed_refs() is called, (unsigned long)-1 means to keep running delayed references while we are able find any, and any other value to run that exact number of delayed references. Typically a specific value other than 0 or -1 is used when flushing space to try to release a certain amount of bytes for a ticket. In this case we just simply calculate how many delayed reference heads correspond to a specific amount of bytes, with calc_delayed_refs_nr(). However that only takes into account the space reserved for the reference heads themselves, and does not account for the space reserved for deleting checksums from the csum tree (see add_delayed_ref_head() and update_existing_head_ref()) in case we are going to delete a data extent. This means we may end up running more delayed references than necessary in case we process delayed references for deleting a data extent. So change the logic of btrfs_run_delayed_refs() to take a bytes argument to specify how many bytes of delayed references to run/release, using the special values of 0 to mean all existing delayed references and U64_MAX (or (u64)-1) to keep running delayed references while we can find any. This prevents running more delayed references than necessary, when we have delayed references for deleting data extents, but also makes the upcoming changes/patches simpler and it's preparatory work for them. Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Filipe Manana <fdmanana@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com>
2023-09-08 18:20:34 +01:00
btrfs: qgroup: track metadata relocation COW with simple quota Relocation COWs metadata blocks in two cases for the reloc root: - copying the subvolume root item when creating the reloc root - copying a btree node when there is a COW during relocation In both cases, the resulting btree node hits an abnormal code path with respect to the owner field in its btrfs_header. It first creates the root item for the new objectid, which populates the reloc root id, and it at this point that delayed refs are created. Later, it fully copies the old node into the new node (including the original owner field) which overwrites it. This results in a simple quotas mismatch where we run the delayed ref for the reloc root which has no simple quota effect (reloc root is not an fstree) but when we ultimately delete the node, the owner is the real original fstree and we do free the space. To work around this without tampering with the behavior of relocation, add a parameter to btrfs_add_tree_block that lets the relocation code path specify a different owning root than the "operating" root (in this case, owning root is the real root and the operating root is the reloc root). These can naturally be plumbed into delayed refs that have the same concept. Note that this is a double count in some sense, but a relatively natural one, as there are really two extents, and the old one will be deleted soon. This is consistent with how data relocation extents are accounted by simple quotas. Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Boris Burkov <boris@bur.io> Signed-off-by: David Sterba <dsterba@suse.com>
2023-06-21 17:37:23 -07:00