From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id B1D3B42DEC; Thu, 6 Jul 2023 19:43:08 +0200 (CEST) Received: from mails.dpdk.org (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id 8588F41101; Thu, 6 Jul 2023 19:43:08 +0200 (CEST) Received: from mail-pg1-f181.google.com (mail-pg1-f181.google.com [209.85.215.181]) by mails.dpdk.org (Postfix) with ESMTP id 20E2F40A79 for ; Thu, 6 Jul 2023 19:43:07 +0200 (CEST) Received: by mail-pg1-f181.google.com with SMTP id 41be03b00d2f7-553ad54d3c6so776790a12.1 for ; Thu, 06 Jul 2023 10:43:07 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=networkplumber-org.20221208.gappssmtp.com; s=20221208; t=1688665386; x=1691257386; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:subject:cc:to:from:date:from:to:cc:subject:date :message-id:reply-to; bh=6UaB+GsuZiT+4u3tdhMkC3Q7s1hKY27r/Hwv8CLk0XI=; b=AxswNzzKjm+qxnpfCCXlBPDCk3vJ1S01SiS3ExDtIMYLAmKZevOfSshHM44E9BvHIt XY39bZNtqPScocemwkAfK8WVXyrRaAHG23u+MLWrOvbDTEHiWMOln+KeB+GlY8BabLFb vXV5/z3iHBD8kZ3wwELv2poellQPdOmGsv5XgVKbgbdAwNu6+7ykAq/bSAfnEXrrSwPo +kjo0QCBJve3hkD3x2MlKkO88OwVuQ6zjH4Pd++A8ou9qup+BuCfV8QVx08sH7C2klzo XGcVV6JyLKg/waNQhAjTHRwXiyYfkCRURcGfVUgA/59bFmdA6UPZRX51Fz09qf20lBj4 pqKw== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20221208; t=1688665386; x=1691257386; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:subject:cc:to:from:date:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=6UaB+GsuZiT+4u3tdhMkC3Q7s1hKY27r/Hwv8CLk0XI=; b=G99Ap+nzWdAvVvnRoIuikvoiFeKH8/1z5bqzu1VcG2HfiY5VYK282pRaHgZoOLeDFA l7/gIzeTVCAzkQ4e5KI5pgM+3nFLUbWYoucssqHRYUD/6Y9iZb0eswYpYq7nzT0X2S+k ndlzMqjoYJpS32xffLRkUlnwsKVMFwb3kvWPN4w18AkMLZ0JljPjjKdvEbjwBh8cRjYJ 6NuilkgPU7g/bKDtUcD9eZSu7WNAGKNFDqz6peZO6vJpzx1m2CCmxuuAZEFjy/H9i2mp 8cDP9n2AeXK+dxr/rzOnsC4KhlN8zSdaz3NpRp2aA/4MEU5SomH4xSVDiGy1Wd1xMast vvzg== X-Gm-Message-State: ABy/qLYFcXbb2n6DNaPMzWa20qtbGM1mEm6Jh780oKemkR18l0d9b5tJ U1Gnxw6tPvrjBzNw3WOp2XsmqA== X-Google-Smtp-Source: APBJJlE5NCTi24izNtOsa24x/NRUxdKJNtNMraHx1XyM7FEUW9TLbJyG1/Kl0pVQ9SECVGwnyVv2fg== X-Received: by 2002:a05:6a20:96cb:b0:118:e70:6f7d with SMTP id hq11-20020a056a2096cb00b001180e706f7dmr1714710pzc.10.1688665384723; Thu, 06 Jul 2023 10:43:04 -0700 (PDT) Received: from hermes.local (204-195-120-218.wavecable.com. [204.195.120.218]) by smtp.gmail.com with ESMTPSA id ja22-20020a170902efd600b001b83a575995sm1713005plb.42.2023.07.06.10.43.03 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Thu, 06 Jul 2023 10:43:04 -0700 (PDT) Date: Thu, 6 Jul 2023 10:43:02 -0700 From: Stephen Hemminger To: Dharmik Thakkar Cc: Jerin Jacob , Morten =?UTF-8?B?QnLDuHJ1cA==?= , Bruce Richardson , Honnappa Nagarahalli , dpdk-dev , nd , Ruifeng Wang Subject: Re: [PATCH 0/1] mempool: implement index-based per core cache Message-ID: <20230706104302.4fd9807a@hermes.local> In-Reply-To: References: <20210930172735.2675627-1-dharmik.thakkar@arm.com> <20211224225923.806498-1-dharmik.thakkar@arm.com> <98CBD80474FA8B44BF855DF32C47DC35D86DAD@smartserver.smartshare.dk> <98CBD80474FA8B44BF855DF32C47DC35D86DEA@smartserver.smartshare.dk> <98CBD80474FA8B44BF855DF32C47DC35D86DEF@smartserver.smartshare.dk> MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: quoted-printable X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org On Thu, 13 Jan 2022 05:31:18 +0000 Dharmik Thakkar wrote: > Hi, >=20 > Thank you for your valuable review comments and suggestions! >=20 > I will be sending out a v2 in which I have increased the size of the memp= ool to 32GB by using division by sizeof(uintptr_t). > However, I am seeing ~5% performance degradation with mempool_perf_autote= st (for bulk size of 32) with this change > when compared to the base performance. > Earlier, without this change, I was seeing an improvement of ~13% compare= d to base performance. So, this is a significant degradation. > I would appreciate your review comments on v2. >=20 > Thank you! >=20 > > On Jan 10, 2022, at 12:38 AM, Jerin Jacob wrote: > >=20 > > On Sat, Jan 8, 2022 at 3:07 PM Morten Br=C3=B8rup wrote: =20 > >> =20 > >>> From: Bruce Richardson [mailto:bruce.richardson@intel.com] > >>> Sent: Friday, 7 January 2022 14.51 > >>>=20 > >>> On Fri, Jan 07, 2022 at 12:29:23PM +0100, Morten Br=C3=B8rup wrote: = =20 > >>>>> From: Bruce Richardson [mailto:bruce.richardson@intel.com] > >>>>> Sent: Friday, 7 January 2022 12.16 > >>>>>=20 > >>>>> On Sat, Dec 25, 2021 at 01:16:03AM +0100, Morten Br=C3=B8rup wrote:= =20 > >>>>>>> From: Dharmik Thakkar [mailto:dharmik.thakkar@arm.com] Sent: =20 > >>>>> Friday, 24 =20 > >>>>>>> December 2021 23.59 > >>>>>>>=20 > >>>>>>> Current mempool per core cache implementation stores pointers =20 > >>> to =20 > >>>>> mbufs =20 > >>>>>>> On 64b architectures, each pointer consumes 8B This patch =20 > >>> replaces =20 > >>>>> it =20 > >>>>>>> with index-based implementation, where in each buffer is =20 > >>> addressed =20 > >>>>> by =20 > >>>>>>> (pool base address + index) It reduces the amount of =20 > >>> memory/cache =20 > >>>>>>> required for per core cache > >>>>>>>=20 > >>>>>>> L3Fwd performance testing reveals minor improvements in the =20 > >>> cache =20 > >>>>>>> performance (L1 and L2 misses reduced by 0.60%) with no change =20 > >>> in =20 > >>>>>>> throughput > >>>>>>>=20 > >>>>>>> Micro-benchmarking the patch using mempool_perf_test shows =20 > >>>>> significant =20 > >>>>>>> improvement with majority of the test cases > >>>>>>> =20 > >>>>>>=20 > >>>>>> I still think this is very interesting. And your performance =20 > >>> numbers =20 > >>>>> are =20 > >>>>>> looking good. > >>>>>>=20 > >>>>>> However, it limits the size of a mempool to 4 GB. As previously > >>>>>> discussed, the max mempool size can be increased by multiplying =20 > >>> the =20 > >>>>> index =20 > >>>>>> with a constant. > >>>>>>=20 > >>>>>> I would suggest using sizeof(uintptr_t) as the constant =20 > >>> multiplier, =20 > >>>>> so =20 > >>>>>> the mempool can hold objects of any size divisible by =20 > >>>>> sizeof(uintptr_t). =20 > >>>>>> And it would be silly to use a mempool to hold objects smaller =20 > >>> than =20 > >>>>>> sizeof(uintptr_t). > >>>>>>=20 > >>>>>> How does the performance look if you multiply the index by > >>>>>> sizeof(uintptr_t)? > >>>>>> =20 > >>>>>=20 > >>>>> Each mempool entry is cache aligned, so we can use that if we want = =20 > >>> a =20 > >>>>> bigger > >>>>> multiplier. =20 > >>>>=20 > >>>> Thanks for chiming in, Bruce. > >>>>=20 > >>>> Please also read this discussion about the multiplier: > >>>> http://inbox.dpdk.org/dev/CALBAE1PrQYyOG96f6ECeW1vPF3TOh1h7MZZULiY95= z9xjbRuyA@mail.gmail.com/ > >>>> =20 > >>>=20 > >>> I actually wondered after I had sent the email whether we had indeed = an > >>> option to disable the cache alignment or not! Thanks for pointing out > >>> that > >>> we do. This brings a couple additional thoughts: > >>>=20 > >>> * Using indexes for the cache should probably be a runtime flag rather > >>> than > >>> a build-time one. > >>> * It would seem reasonable to me to disallow use of the indexed-cache > >>> flag > >>> and the non-cache aligned flag simultaneously. > >>> * On the offchance that that restriction is unacceptable, then we can > >>> make things a little more complicated by doing a runtime computation > >>> of > >>> the "index-shiftwidth" to use. > >>>=20 > >>> Overall, I think defaulting to cacheline shiftwidth and disallowing > >>> index-based addressing when using unaligned buffers is simplest and > >>> easiest > >>> unless we can come up with a valid usecase for needing more than that. > >>>=20 > >>> /Bruce =20 > >>=20 > >> This feature is a performance optimization. > >>=20 > >> With that in mind, it should not introduce function pointers or simila= r run-time checks or in the fast path, to determine what kind of cache to u= se per mempool. And if an index multiplier is implemented, it should be a c= ompile time constant, probably something between sizeof(uintptr_t) or RTE_M= EMPOOL_ALIGN (=3DRTE_CACHE_LINE_SIZE). > >>=20 > >> The patch comes with a tradeoff between better performance and limited= mempool size, and possibly some limitations regarding very small objects t= hat are not cache line aligned to avoid wasting memory (RTE_MEMPOOL_POPULAT= E_F_ALIGN_OBJ). > >>=20 > >> With no multiplier, the only tradeoff is that the mempool size is limi= ted to 4 GB. > >>=20 > >> If the multiplier is small (i.e. 8 bytes) the only tradeoff is that th= e mempool size is limited to 32 GB. (And a waste of memory for objects smal= ler than 8 byte; but I don't think anyone would use a mempool to hold objec= ts smaller than 8 byte.) > >>=20 > >> If the multiplier is larger (i.e. 64 bytes cache line size), the mempo= ol size is instead limited to 256 GB, but RTE_MEMPOOL_POPULATE_F_ALIGN_OBJ = has no effect. > >>=20 > >> Note: 32 bit platforms have no benefit from this patch: The pointer al= ready only uses 4 bytes, so replacing the pointer with a 4 byte index makes= no difference. > >>=20 > >>=20 > >> Since this feature is a performance optimization only, and doesn't pro= vide any new features, I don't mind it being a compile time option. > >>=20 > >> If this feature is a compile time option, and the mempool library is c= ompiled with the large multiplier, then RTE_MEMPOOL_POPULATE_F_ALIGN_OBJ co= uld be made undefined in the public header file, so compilation of applicat= ions using the flag will fail. And rte_mempool_create() could RTE_ASSERT() = that RTE_MEMPOOL_POPULATE_F_ALIGN_OBJ is not set in its flags parameter, or= emit a warning about the flag being ignored. Obviously, rte_mempool_create= () should also RTE_ASSERT() that the mempool is not larger than the library= supports, possibly emitting a message that the mempool library should be b= uilt without this feature to support the larger mempool. > >>=20 > >> Here is another thought: If only exotic applications use mempools larg= er than 32 GB, this would be a generally acceptable limit, and DPDK should = use index-based cache as default, making the opposite (i.e. pointer-based c= ache) a compile time option instead. A similar decision was recently made f= or limiting the RTE_MAX_LCORE default. > >>=20 > >>=20 > >> Although DPDK is moving away from compile time options in order to bet= ter support Linux distros, there should be a general exception for performa= nce and memory optimizations. Otherwise, network appliance vendors will inh= erit the increasing amount of DPDK bloat, and we (network appliance vendors= ) will eventually be forced to fork DPDK to get rid of the bloat and achiev= e the goals originally intended by DPDK. =20 > >=20 > > Agree with Morten's view on this. > > =20 > >> If anyone disagrees with the principle about a general exception for p= erformance and memory optimizations, I would like to pass on the decision t= o the Techboard! > >> =20 NAK Having compile time stuff like this means one side or the other is not test= ed by CI infrastructure. There never was sufficient justification, and lots o= f objections. Dropping the patch.