[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <ff610f19-7303-f583-4e22-e526f314aaa9@gmx.com>
Date: Mon, 27 Feb 2023 19:46:00 +0800
From: Qu Wenruo <quwenruo.btrfs@....com>
To: Filipe Manana <fdmanana@...nel.org>
Cc: Chris Mason <clm@...com>, Josef Bacik <josef@...icpanda.com>,
David Sterba <dsterba@...e.com>, Tejun Heo <tj@...nel.org>,
Lai Jiangshan <jiangshanlai@...il.com>,
Filipe Manana <fdmanana@...e.com>,
Linux Btrfs Mailing List <linux-btrfs@...r.kernel.org>,
Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
Linux Fsdevel Mailing List <linux-fsdevel@...r.kernel.org>
Subject: Re: [RFC PATCH v1 0/6] Introducing `wq_cpu_set` mount option for
btrfs
On 2023/2/27 19:02, Filipe Manana wrote:
> On Sun, Feb 26, 2023 at 4:31 PM Ammar Faizi <ammarfaizi2@...weeb.org> wrote:
>>
>> Hi,
>>
>> This is an RFC patchset that introduces the `wq_cpu_set` mount option.
>> This option lets the user specify a CPU set that the Btrfs workqueues
>> will use.
>>
>> Btrfs workqueues can slow sensitive user tasks down because they can use
>> any online CPU to perform heavy workloads on an SMP system. Add a mount
>> option to isolate the Btrfs workqueues to a set of CPUs. It is helpful
>> to avoid sensitive user tasks being preempted by Btrfs heavy workqueues.
>>
>> This option is similar to the taskset bitmask except that the comma
>> separator is replaced with a dot. The reason for this is that the mount
>> option parser uses commas to separate mount options.
>>
>> Figure (the CPU usage when `wq_cpu_set` is used VS when it is not):
>> https://gist.githubusercontent.com/ammarfaizi2/a10f8073e58d1712c1ed49af83ae4ad1/raw/a4f7cbc4eb163db792a669d570ff542495e8c704/wq_cpu_set.png
>
> I haven't read the patchset.
>
> It's great that it reduces CPU usage.
> But does it also provide other performance benefits, like lower
> latency or higher throughput for some workloads? Or using less CPU
> also affects negatively in those other aspects?
So far it looks like to just set CPU masks for each workqueue.
Thus if it's reducing CPU usage, it also takes longer time to finish the
workload (compression,csum calculation etc).
Thanks,
Qu
>
> Thanks.
>
>>
>> A simple stress testing:
>>
>> 1. Open htop.
>> 2. Open a new terminal.
>> 3. Mount and perform a heavy workload on the mounted Btrfs filesystem.
>>
>> ## Test without wq_cpu_set
>> sudo mount -t btrfs -o rw,compress-force=zstd:15,commit=1500 /dev/sda2 hdd/a;
>> cp -rf /path/folder_with_many_large_files/ hdd/a/test;
>> sync; # See the CPU usage in htop.
>> sudo umount hdd/a;
>>
>> ## Test wq_cpu_set
>> sudo mount -t btrfs -o rw,compress-force=zstd:15,commit=1500,wq_cpu_set=0.4.1.5 /dev/sda2 hdd/a;
>> cp -rf /path/folder_with_many_large_files/ hdd/a/test;
>> sync; # See the CPU usage in htop.
>> sudo umount hdd/a;
>>
>> Signed-off-by: Ammar Faizi <ammarfaizi2@...weeb.org>
>> ---
>>
>> Ammar Faizi (6):
>> workqueue: Add set_workqueue_cpumask() helper function
>> btrfs: Change `mount_opt` type in `struct btrfs_fs_info` to `u64`
>> btrfs: Create btrfs CPU set struct and helpers
>> btrfs: Add wq_cpu_set=%s mount option
>> btrfs: Adjust the default thread pool size when `wq_cpu_set` option is used
>> btrfs: Add `BTRFS_DEFAULT_MAX_THREAD_POOL_SIZE` macro
>>
>> fs/btrfs/async-thread.c | 51 ++++++++++++++++++++
>> fs/btrfs/async-thread.h | 3 ++
>> fs/btrfs/disk-io.c | 6 ++-
>> fs/btrfs/fs.c | 97 +++++++++++++++++++++++++++++++++++++++
>> fs/btrfs/fs.h | 12 ++++-
>> fs/btrfs/super.c | 83 +++++++++++++++++++++++++++++++++
>> include/linux/workqueue.h | 3 ++
>> kernel/workqueue.c | 19 ++++++++
>> 8 files changed, 271 insertions(+), 3 deletions(-)
>>
>>
>> base-commit: 2fcd07b7ccd5fd10b2120d298363e4e6c53ccf9c
>> --
>> Ammar Faizi
>>
Powered by blists - more mailing lists