From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from lists.zx2c4.com (lists.zx2c4.com [165.227.139.114]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id BF434C4167B for ; Mon, 4 Dec 2023 18:11:00 +0000 (UTC) Received: by lists.zx2c4.com (ZX2C4 Mail Server) with ESMTP id 8df89acf; Mon, 4 Dec 2023 18:07:13 +0000 (UTC) Received: from mail-pf1-x430.google.com (mail-pf1-x430.google.com [2607:f8b0:4864:20::430]) by lists.zx2c4.com (ZX2C4 Mail Server) with ESMTPS id 068a753c (TLSv1.3:TLS_AES_256_GCM_SHA384:256:NO) for ; Mon, 4 Dec 2023 18:07:11 +0000 (UTC) Received: by mail-pf1-x430.google.com with SMTP id d2e1a72fcca58-6ce416a74d5so907361b3a.2 for ; Mon, 04 Dec 2023 10:07:11 -0800 (PST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20230601; t=1701713230; x=1702318030; darn=lists.zx2c4.com; h=in-reply-to:content-disposition:mime-version:references:message-id :subject:cc:to:from:date:sender:from:to:cc:subject:date:message-id :reply-to; bh=vnKGL9rVmaXJ0k7UQTGGb8qgGBw3yaFYXulmlj0ClVk=; b=bCixj4xeznuIN0t+tS1vfQFivkZpc+X1W09ZpQpqdUby7UqUF+ECEpsIZ9tZ5+1RTD OKFvw8DPoyJjLYHJuvpA3d2hvaP//NFS9TgT7UkSGWCAX+Y1g8I7hLFj/37f+7AC+f90 UJjdSLbS7Wk8zo8QjqoqB37jhLBXeSNQxS//8Ih53Zzj3pB5Yv6QIP0cOiys2S256qKT 5mf44Ho5/u5/xFzaSTg5+mIEZQeTGwZYGny07JTnY/gRgRzPNrsp4iJeDUT5PXSDC6+6 CO7kBJMZ0AfA6KUkD9LzhWLovXMS8Ur5tyOKtcj22sUWWihaPuabSBSOQIAPxMrpq+OS rqAw== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20230601; t=1701713230; x=1702318030; h=in-reply-to:content-disposition:mime-version:references:message-id :subject:cc:to:from:date:sender:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=vnKGL9rVmaXJ0k7UQTGGb8qgGBw3yaFYXulmlj0ClVk=; b=SHf3oaqB8wga/OHiPOYSCus8Y73akzFfiVASa70YH/uIXe3nPLMk44T0HWi0FLbowi SbZIqeASSldHDgjxPNpSiY5ByR2oeuL9RnE0ZAT+1w2p0jWu7B9IkVhsX6jCeiPe0A7c MNiadIbBvdWjCMkH1/7wobXgCyFsLtjeerHWg3X4LYKlxwh0zGHPuskxRkFFRJf/YvZq lc/mGL4OL3dSwE6sifUfKtPoecbIeTkCJmjix5N4JRGcM1oLn97r3K7eTWoGOs7lDfCv 7f2EGSJe0CvsUpfrXcTP9JdVWUKitut62j4oDmxNlP2OQ85zZ8QJ0lA00y3pFiI96APQ 9etg== X-Gm-Message-State: AOJu0YyZvC6VnDeIOC+7sVR+L//pvwVzXURmKEtKljbBWW4tFxLWsf6u JN2VGHOF2sFvjKqEmCGjhg4= X-Google-Smtp-Source: AGHT+IGDIGeeBa7UWPj4E4w2SIGHBjZgGz8oxGqwAWzNb+KD7mpnEvIno4bt3NIHORxil3wMtNwC4w== X-Received: by 2002:a05:6a20:1448:b0:18c:374c:6e64 with SMTP id a8-20020a056a20144800b0018c374c6e64mr27716780pzi.36.1701713229690; Mon, 04 Dec 2023 10:07:09 -0800 (PST) Received: from localhost ([2620:10d:c090:400::4:27ef]) by smtp.gmail.com with ESMTPSA id u2-20020a056a00158200b006cdd507ca2esm7943470pfk.167.2023.12.04.10.07.08 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 04 Dec 2023 10:07:09 -0800 (PST) Date: Mon, 4 Dec 2023 08:07:07 -1000 From: Tejun Heo To: Naohiro Aota Cc: Lai Jiangshan , "linux-kernel@vger.kernel.org" , "linux-btrfs@vger.kernel.org" , "ceph-devel@vger.kernel.org" , "cgroups@vger.kernel.org" , "coreteam@netfilter.org" , "dm-devel@lists.linux.dev" , "dri-devel@lists.freedesktop.org" , "gfs2@lists.linux.dev" , "intel-gfx@lists.freedesktop.org" , "iommu@lists.linux.dev" , "linux-arm-kernel@lists.infradead.org" , "linux-bcachefs@vger.kernel.org" , "linux-block@vger.kernel.org" , "linux-cachefs@redhat.com" , "linux-cifs@vger.kernel.org" , "linux-crypto@vger.kernel.org" , "linux-erofs@lists.ozlabs.org" , "linux-f2fs-devel@lists.sourceforge.net" , "linux-fscrypt@vger.kernel.org" , "linux-media@vger.kernel.org" , "linux-mediatek@lists.infradead.org" , "linux-mm@kvack.org" , "linux-mmc@vger.kernel.org" , "linux-nfs@vger.kernel.org" , "linux-nvme@lists.infradead.org" , "linux-raid@vger.kernel.org" , "linux-rdma@vger.kernel.org" , "linux-remoteproc@vger.kernel.org" , "linux-scsi@vger.kernel.org" , "linux-trace-kernel@vger.kernel.org" , "linux-usb@vger.kernel.org" , "linux-wireless@vger.kernel.org" , "linux-xfs@vger.kernel.org" , "nbd@other.debian.org" , "netdev@vger.kernel.org" , "ntb@lists.linux.dev" , "open-iscsi@googlegroups.com" , "oss-drivers@corigine.com" , "platform-driver-x86@vger.kernel.org" , "samba-technical@lists.samba.org" , "target-devel@vger.kernel.org" , "virtualization@lists.linux.dev" , "wireguard@lists.zx2c4.com" Subject: Re: Performance drop due to alloc_workqueue() misuse and recent change Message-ID: References: MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: X-BeenThere: wireguard@lists.zx2c4.com X-Mailman-Version: 2.1.30rc1 Precedence: list List-Id: Development discussion of WireGuard List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: wireguard-bounces@lists.zx2c4.com Sender: "WireGuard" Hello, On Mon, Dec 04, 2023 at 04:03:47PM +0000, Naohiro Aota wrote: > Recently, commit 636b927eba5b ("workqueue: Make unbound workqueues to use > per-cpu pool_workqueues") changed WQ_UNBOUND workqueue's behavior. It > changed the meaning of alloc_workqueue()'s max_active from an upper limit > imposed per NUMA node to a limit per CPU. As a result, massive number of > workers can be running at the same time, especially if the workqueue user > thinks the max_active is a global limit. > > Actually, it is already written it is per-CPU limit in the documentation > before the commit. However, several callers seem to misuse max_active, > maybe thinking it is a global limit. It is an unexpected behavior change > for them. Right, and the behavior has been like that for a very long time and there was no other way to achieve reasonable level of concurrency, so the current situation is expected. > For example, these callers set max_active = num_online_cpus(), which is a > suspicious limit applying to per-CPU. This config means we can have nr_cpu > * nr_cpu active tasks working at the same time. Yeah, that sounds like a good indicator. > fs/f2fs/data.c: sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq", > fs/f2fs/data.c- WQ_UNBOUND | WQ_HIGHPRI, > fs/f2fs/data.c- num_online_cpus()); > > fs/crypto/crypto.c: fscrypt_read_workqueue = alloc_workqueue("fscrypt_read_queue", > fs/crypto/crypto.c- WQ_UNBOUND | WQ_HIGHPRI, > fs/crypto/crypto.c- num_online_cpus()); > > fs/verity/verify.c: fsverity_read_workqueue = alloc_workqueue("fsverity_read_queue", > fs/verity/verify.c- WQ_HIGHPRI, > fs/verity/verify.c- num_online_cpus()); > > drivers/crypto/hisilicon/qm.c: qm->wq = alloc_workqueue("%s", WQ_HIGHPRI | WQ_MEM_RECLAIM | > drivers/crypto/hisilicon/qm.c- WQ_UNBOUND, num_online_cpus(), > drivers/crypto/hisilicon/qm.c- pci_name(qm->pdev)); > > block/blk-crypto-fallback.c: blk_crypto_wq = alloc_workqueue("blk_crypto_wq", > block/blk-crypto-fallback.c- WQ_UNBOUND | WQ_HIGHPRI | > block/blk-crypto-fallback.c- WQ_MEM_RECLAIM, num_online_cpus()); > > drivers/md/dm-crypt.c: cc->crypt_queue = alloc_workqueue("kcryptd/%s", > drivers/md/dm-crypt.c- WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM | WQ_UNBOUND, > drivers/md/dm-crypt.c- num_online_cpus(), devname); Most of these work items are CPU bound but not completley so. e.g. kcrypt_crypt_write_continue() does wait_for_completion(), so setting max_active to 1 likely isn't what they want either. They mostly want some reasonable system-wide concurrency limit w.r.t. the CPU count while keeping some level of flexibility in terms of task placement. The previous max_active wasn't great for this because its meaning changed depending on the number of nodes. Now, the meaning doesn't change but it's not really useful for the above purpose. It's only useful for avoiding melting the system completely. One way to go about it is to declare that concurrency level management for unbound workqueue is on users but that seems not ideal given many use cases would want it anyway. Let me think it over but I think the right way to go about it is going the other direction - ie. making max_active apply to the whole system regardless of the number of nodes / ccx's / whatever. > Furthermore, the change affects performance in a certain case. > > Btrfs creates several WQ_UNBOUND workqueues with a default max_active = > min(NRCPUS + 2, 8). As my machine has 96 CPUs with NUMA disabled, this > max_active config allows running over 700 active works. Before the commit, > it is limited to 8 if NUMA is disabled or limited to 16 if NUMA nodes is 2. > > I reverted the workqueue code back to before the commit, and I ran the > following fio command on RAID0 btrfs on 6 SSDs. > > fio --group_reporting --eta=always --eta-interval=30s --eta-newline=30s \ > --rw=write --fallocate=none \ > --direct=1 --ioengine=libaio --iodepth=32 \ > --filesize=100G \ > --blocksize=64k \ > --time_based --runtime=300s \ > --end_fsync=1 \ > --directory=${MNT} \ > --name=writer --numjobs=32 > > By changing workqueue's max_active, the result varies. > > - wq max_active=8 (intended limit by btrfs?) > WRITE: bw=2495MiB/s (2616MB/s), 2495MiB/s-2495MiB/s (2616MB/s-2616MB/s), io=753GiB (808GB), run=308953-308953msec > - wq max_active=16 (actual limit on 2 NUMA nodes setup) > WRITE: bw=1736MiB/s (1820MB/s), 1736MiB/s-1736MiB/s (1820MB/s-1820MB/s), io=670GiB (720GB), run=395532-395532msec > - wq max_active=768 (simulating current limit) > WRITE: bw=1276MiB/s (1338MB/s), 1276MiB/s-1276MiB/s (1338MB/s-1338MB/s), io=375GiB (403GB), run=300984-300984msec > > The current performance is slower than the previous limit (max_active=16) > by 27%, or it is 50% slower than the intended limit. The performance drop > might be due to contention of the btrfs-endio-write works. There are over > 700 kworker instances were created and 100 works are on the 'D' state > competing for a lock. > > More specifically, I tested the same workload on the commit. > > - At commit 636b927eba5b ("workqueue: Make unbound workqueues to use per-cpu pool_workqueues") > WRITE: bw=1191MiB/s (1249MB/s), 1191MiB/s-1191MiB/s (1249MB/s-1249MB/s), io=350GiB (376GB), run=300714-300714msec > - At the previous commit = 4cbfd3de73 ("workqueue: Call wq_update_unbound_numa() on all CPUs in NUMA node on CPU hotplug") > WRITE: bw=1747MiB/s (1832MB/s), 1747MiB/s-1747MiB/s (1832MB/s-1832MB/s), io=748GiB (803GB), run=438134-438134msec > > So, it is -31.8% performance down with the commit. > > In summary, we misuse max_active, considering it is a global limit. And, > the recent commit introduced a huge performance drop in some cases. We > need to review alloc_workqueue() usage to check if its max_active setting > is proper or not. Thanks a lot for the report. I think it's a lot more reasonable to assume that max_active is global for unbound workqueues. The current workqueue behavior is not very intuitive or useful. I'll try to find something more reasonable. Thanks for the report and analysis. Much appreciated. Thanks. -- tejun