From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Dumazet Subject: Re: [net-next PATCH 1/6] net: Split netdev_alloc_frag into __alloc_page_frag and add __napi_alloc_frag Date: Wed, 10 Dec 2014 08:02:08 -0800 Message-ID: <1418227328.27198.25.camel@edumazet-glaptop2.roam.corp.google.com> References: <20141210033902.2114.68658.stgit@ahduyck-vm-fedora20> <20141210034042.2114.29360.stgit@ahduyck-vm-fedora20> Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Cc: netdev@vger.kernel.org, ast@plumgrid.com, davem@davemloft.net, brouer@redhat.com To: Alexander Duyck Return-path: Received: from mail-ig0-f174.google.com ([209.85.213.174]:39939 "EHLO mail-ig0-f174.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932116AbaLJQCL (ORCPT ); Wed, 10 Dec 2014 11:02:11 -0500 Received: by mail-ig0-f174.google.com with SMTP id hn15so6679354igb.7 for ; Wed, 10 Dec 2014 08:02:10 -0800 (PST) In-Reply-To: <20141210034042.2114.29360.stgit@ahduyck-vm-fedora20> Sender: netdev-owner@vger.kernel.org List-ID: On Tue, 2014-12-09 at 19:40 -0800, Alexander Duyck wrote: > I also took the opportunity to refactor the core bits that were placed in > __alloc_page_frag. First I updated the allocation to do either a 32K > allocation or an order 0 page. This is based on the changes in commmit > d9b2938aa where it was found that latencies could be reduced in case of > failures. GFP_KERNEL and GFP_ATOMIC allocation constraints are quite different. I have no idea how expensive it is to attempt order-3, order-2, order-1 allocations with GFP_ATOMIC. I did an interesting experiment on mlx4 driver, allocating the pages needed to store the fragments, using a small layer before the alloc_page() that is normally used : - Attempt order-9 allocations, and use split_page() to give the individual pages. Boost in performance is 10% on TCP bulk receive, because of less TLB misses. With huge amount of memory these days, alloc_page() tend to give pages spread all over memory, with poor TLB locality. With this strategy, a 1024 RX ring is backed by 2 huge pages only.