From mboxrd@z Thu Jan 1 00:00:00 1970 From: Waiman Long Subject: Re: [PATCH v9 0/5] Add NUMA-awareness to qspinlock Date: Wed, 22 Jan 2020 12:24:58 -0500 Message-ID: <4e15fa1d-9540-3274-502a-4195a0d46f63@redhat.com> References: <20200115035920.54451-1-alex.kogan@oracle.com> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: quoted-printable Return-path: In-Reply-To: Content-Language: en-US Sender: linux-kernel-owner@vger.kernel.org To: Lihao Liang , Alex Kogan Cc: linux@armlinux.org.uk, peterz@infradead.org, mingo@redhat.com, will.deacon@arm.com, arnd@arndb.de, linux-arch@vger.kernel.org, linux-arm-kernel@lists.infradead.org, linux-kernel@vger.kernel.org, tglx@linutronix.de, bp@alien8.de, hpa@zytor.com, x86@kernel.org, guohanjun@huawei.com, jglauber@marvell.com, dave.dice@oracle.com, steven.sistare@oracle.com, daniel.m.jordan@oracle.com, Will Deacon List-Id: linux-arch.vger.kernel.org On 1/22/20 6:45 AM, Lihao Liang wrote: > Hi Alex, > > On Wed, Jan 22, 2020 at 10:28 AM Alex Kogan wro= te: >> Summary >> ------- >> >> Lock throughput can be increased by handing a lock to a waiter on the >> same NUMA node as the lock holder, provided care is taken to avoid >> starvation of waiters on other NUMA nodes. This patch introduces CNA >> (compact NUMA-aware lock) as the slow path for qspinlock. It is >> enabled through a configuration option (NUMA_AWARE_SPINLOCKS). >> > Thanks for your patches. The experimental results look promising! > > I understand that the new CNA qspinlock uses randomization to achieve > long-term fairness, and provides the numa_spinlock_threshold parameter > for users to tune. As Linux runs extremely diverse workloads, it is not > clear how randomization affects its fairness, and how users with > different requirements are supposed to tune this parameter. > > To this end, Will and I consider it beneficial to be able to answer the > following question: > > With different values of numa_spinlock_threshold and > SHUFFLE_REDUCTION_PROB_ARG, how long do threads running on different > sockets have to wait to acquire the lock? This is particularly relevant > in high contention situations when new threads keep arriving on the sam= e > socket as the lock holder. > > In this email, I try to provide some formal analysis to address this > question. Let's assume the probability for the lock to stay on the > same socket is *at least* p, which corresponds to the probability for > the function probably(unsigned int num_bits) in the patch to return *fa= lse*, > where SHUFFLE_REDUCTION_PROB_ARG is passed as the value of num_bits to = the > function. That is not strictly true from my understanding of the code. The probably() function does not come into play if a secondary queue is present. Also calling cna_scan_main_queue() doesn't guarantee that a waiter in the same node can be found. So the simple mathematical analysis isn't that applicable in this case. One will have to do an actual simulation to find out what the actual behavior will be. The comment in the code states that: /* =C2=A0* Controls the probability for enabling the scan of the main queue = when =C2=A0* the secondary queue is empty. The chosen value reduces the amount= of =C2=A0* unnecessary shuffling of threads between the two waiting queues w= hen =C2=A0* the contention is low, while responding fast enough and enabling =C2=A0* the shuffling when the contention is high. =C2=A0*/ #define SHUFFLE_REDUCTION_PROB_ARG=C2=A0 (7) Cheers, Longman From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from us-smtp-delivery-1.mimecast.com ([205.139.110.120]:51466 "EHLO us-smtp-1.mimecast.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1725883AbgAVRZL (ORCPT ); Wed, 22 Jan 2020 12:25:11 -0500 Subject: Re: [PATCH v9 0/5] Add NUMA-awareness to qspinlock References: <20200115035920.54451-1-alex.kogan@oracle.com> From: Waiman Long Message-ID: <4e15fa1d-9540-3274-502a-4195a0d46f63@redhat.com> Date: Wed, 22 Jan 2020 12:24:58 -0500 MIME-Version: 1.0 In-Reply-To: Content-Type: text/plain; charset=utf-8 Content-Language: en-US Content-Transfer-Encoding: quoted-printable Sender: linux-arch-owner@vger.kernel.org List-ID: To: Lihao Liang , Alex Kogan Cc: linux@armlinux.org.uk, peterz@infradead.org, mingo@redhat.com, will.deacon@arm.com, arnd@arndb.de, linux-arch@vger.kernel.org, linux-arm-kernel@lists.infradead.org, linux-kernel@vger.kernel.org, tglx@linutronix.de, bp@alien8.de, hpa@zytor.com, x86@kernel.org, guohanjun@huawei.com, jglauber@marvell.com, dave.dice@oracle.com, steven.sistare@oracle.com, daniel.m.jordan@oracle.com, Will Deacon Message-ID: <20200122172458.vtFzlBncBzSQ5R-WnsVAvyETf9luvBsOhyMs0yOro8E@z> On 1/22/20 6:45 AM, Lihao Liang wrote: > Hi Alex, > > On Wed, Jan 22, 2020 at 10:28 AM Alex Kogan wro= te: >> Summary >> ------- >> >> Lock throughput can be increased by handing a lock to a waiter on the >> same NUMA node as the lock holder, provided care is taken to avoid >> starvation of waiters on other NUMA nodes. This patch introduces CNA >> (compact NUMA-aware lock) as the slow path for qspinlock. It is >> enabled through a configuration option (NUMA_AWARE_SPINLOCKS). >> > Thanks for your patches. The experimental results look promising! > > I understand that the new CNA qspinlock uses randomization to achieve > long-term fairness, and provides the numa_spinlock_threshold parameter > for users to tune. As Linux runs extremely diverse workloads, it is not > clear how randomization affects its fairness, and how users with > different requirements are supposed to tune this parameter. > > To this end, Will and I consider it beneficial to be able to answer the > following question: > > With different values of numa_spinlock_threshold and > SHUFFLE_REDUCTION_PROB_ARG, how long do threads running on different > sockets have to wait to acquire the lock? This is particularly relevant > in high contention situations when new threads keep arriving on the sam= e > socket as the lock holder. > > In this email, I try to provide some formal analysis to address this > question. Let's assume the probability for the lock to stay on the > same socket is *at least* p, which corresponds to the probability for > the function probably(unsigned int num_bits) in the patch to return *fa= lse*, > where SHUFFLE_REDUCTION_PROB_ARG is passed as the value of num_bits to = the > function. That is not strictly true from my understanding of the code. The probably() function does not come into play if a secondary queue is present. Also calling cna_scan_main_queue() doesn't guarantee that a waiter in the same node can be found. So the simple mathematical analysis isn't that applicable in this case. One will have to do an actual simulation to find out what the actual behavior will be. The comment in the code states that: /* =C2=A0* Controls the probability for enabling the scan of the main queue = when =C2=A0* the secondary queue is empty. The chosen value reduces the amount= of =C2=A0* unnecessary shuffling of threads between the two waiting queues w= hen =C2=A0* the contention is low, while responding fast enough and enabling =C2=A0* the shuffling when the contention is high. =C2=A0*/ #define SHUFFLE_REDUCTION_PROB_ARG=C2=A0 (7) Cheers, Longman