c/mirror_zfs
1
0
Fork 0
mirror of https://git.proxmox.com/git/mirror_zfs.git synced 2026-05-22 18:40:43 +03:00
Code Issues Packages Projects Releases Wiki Activity

Improve ZFS objset sync parallelism

Browse Source 
As part of transaction group commit, dsl_pool_sync() sequentially calls
dsl_dataset_sync() for each dirty dataset, which subsequently calls
dmu_objset_sync().  dmu_objset_sync() in turn uses up to 75% of CPU
cores to run sync_dnodes_task() in taskq threads to sync the dirty
dnodes (files).

There are two problems:

1. Each ZVOL in a pool is a separate dataset/objset having a single
   dnode.  This means the objsets are synchronized serially, which
   leads to a bottleneck of ~330K blocks written per second per pool.

2. In the case of multiple dirty dnodes/files on a dataset/objset on a
   big system they will be sync'd in parallel taskq threads. However,
   it is inefficient to to use 75% of CPU cores of a big system to do
   that, because of (a) bottlenecks on a single write issue taskq, and
   (b) allocation throttling.  In addition, if not for the allocation
   throttling sorting write requests by bookmarks (logical address),
   writes for different files may reach space allocators interleaved,
   leading to unwanted fragmentation.

The solution to both problems is to always sync no more and (if
possible) no fewer dnodes at the same time than there are allocators
the pool.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Edmund Nadolski <edmund.nadolski@ixsystems.com>
Closes #15197
This commit is contained in:
ednadolski-ix
2023-11-06 11:38:42 -07:00
committed by GitHub
parent 0527774066
commit 3bd4df3841
18 changed files with 513 additions and 118 deletions
Download Patch File Download Diff File Expand all files Collapse all files
man/man4/zfs.4
+15 -7
View File
@@ -496,6 +496,13 @@ If we have less than this amount of free space,
most ZPL operations (e.g. write, create) will return
.Sy ENOSPC .
.
.It Sy spa_num_allocators Ns = Ns Sy 4 Pq int
Determines the number of block alloctators to use per spa instance.
Capped by the number of actual CPUs in the system.
.Pp
Note that setting this value too high could result in performance
degredation and/or excess fragmentation.
.
.It Sy spa_upgrade_errlog_limit Ns = Ns Sy 0 Pq uint
Limits the number of on-disk error log entries that will be converted to the
new format when enabling the
@@ -1974,13 +1981,6 @@ and may need to load new metaslabs to satisfy these allocations.
.It Sy zfs_sync_pass_rewrite Ns = Ns Sy 2 Pq uint
Rewrite new block pointers starting in this pass.
.
.It Sy zfs_sync_taskq_batch_pct Ns = Ns Sy 75 Ns % Pq int
This controls the number of threads used by
.Sy dp_sync_taskq .
The default value of
.Sy 75%
will create a maximum of one thread per CPU.
.
.It Sy zfs_trim_extent_bytes_max Ns = Ns Sy 134217728 Ns B Po 128 MiB Pc Pq uint
Maximum size of TRIM command.
Larger ranges will be split into chunks no larger than this value before
@@ -2265,6 +2265,14 @@ If
.Sy 0 ,
generate a system-dependent value close to 6 threads per taskq.
.
.It Sy zio_taskq_wr_iss_ncpus Ns = Ns Sy 0 Pq uint
Determines the number of CPUs to run write issue taskqs.
.Pp
When 0 (the default), the value to use is computed internally
as the number of actual CPUs in the system divided by the
.Sy spa_num_allocators
value.
.
.It Sy zvol_inhibit_dev Ns = Ns Sy 0 Ns | Ns 1 Pq uint
Do not create zvol device nodes.
This may slightly improve startup time on