From: Robin Holt <holt@sgi.com> To: Ingo Molnar <mingo@kernel.org> Cc: Robin Holt <holt@sgi.com>, Borislav Petkov <bp@alien8.de>, Robert Richter <rric@kernel.org>, "H. Peter Anvin" <hpa@zytor.com>, Nate Zimmer <nzimmer@sgi.com>, Linux Kernel <linux-kernel@vger.kernel.org>, Linux MM <linux-mm@kvack.org>, Rob Landley <rob@landley.net>, Mike Travis <travis@sgi.com>, Daniel J Blueman <daniel@numascale-asia.com>, Andrew Morton <akpm@linux-foundation.org>, Greg KH <gregkh@linuxfoundation.org>, Yinghai Lu <yinghai@kernel.org>, Mel Gorman <mgorman@suse.de>, Peter Zijlstra <a.p.zijlstra@chello.nl> Subject: Re: [RFC 0/4] Transparent on-demand struct page initialization embedded in the buddy allocator Date: Mon, 15 Jul 2013 10:16:23 -0500 [thread overview] Message-ID: <20130715151623.GB3421@sgi.com> (raw) In-Reply-To: <20130712082756.GA4328@gmail.com> On Fri, Jul 12, 2013 at 10:27:56AM +0200, Ingo Molnar wrote: > > * Robin Holt <holt@sgi.com> wrote: > > > [...] > > > > With this patch, we did boot a 16TiB machine. Without the patches, the > > v3.10 kernel with the same configuration took 407 seconds for > > free_all_bootmem. With the patches and operating on 2MiB pages instead > > of 1GiB, it took 26 seconds so performance was improved. I have no feel > > for how the 1GiB chunk size will perform. > > That's pretty impressive. And WRONG! That is a 15x speedup in the freeing of memory at the free_all_bootmem point. That is _NOT_ the speedup from memmap_init_zone. I forgot to take that into account as Nate pointed out this morning in a hallway discussion. Before, on the 16TiB machine, memmap_init_zone took 1152 seconds. After, it took 50. If it were a straight 1/512th, we would have expected that 1152 to be something more on the line of 2-3 so there is still significant room for improvement. Sorry for the confusion. > It's still a 15x speedup instead of a 512x speedup, so I'd say there's > something else being the current bottleneck, besides page init > granularity. > > Can you boot with just a few gigs of RAM and stuff the rest into hotplug > memory, and then hot-add that memory? That would allow easy profiling of > remaining overhead. Nate and I will be working on other things for the next few hours hoping there is a better answer to the first question we asked about there being a way to test a page other than comparing against all zeroes to see if it has been initialized. Thanks, Robin
WARNING: multiple messages have this Message-ID (diff)
From: Robin Holt <holt@sgi.com> To: Ingo Molnar <mingo@kernel.org> Cc: Robin Holt <holt@sgi.com>, Borislav Petkov <bp@alien8.de>, Robert Richter <rric@kernel.org>, "H. Peter Anvin" <hpa@zytor.com>, Nate Zimmer <nzimmer@sgi.com>, Linux Kernel <linux-kernel@vger.kernel.org>, Linux MM <linux-mm@kvack.org>, Rob Landley <rob@landley.net>, Mike Travis <travis@sgi.com>, Daniel J Blueman <daniel@numascale-asia.com>, Andrew Morton <akpm@linux-foundation.org>, Greg KH <gregkh@linuxfoundation.org>, Yinghai Lu <yinghai@kernel.org>, Mel Gorman <mgorman@suse.de>, Peter Zijlstra <a.p.zijlstra@chello.nl> Subject: Re: [RFC 0/4] Transparent on-demand struct page initialization embedded in the buddy allocator Date: Mon, 15 Jul 2013 10:16:23 -0500 [thread overview] Message-ID: <20130715151623.GB3421@sgi.com> (raw) In-Reply-To: <20130712082756.GA4328@gmail.com> On Fri, Jul 12, 2013 at 10:27:56AM +0200, Ingo Molnar wrote: > > * Robin Holt <holt@sgi.com> wrote: > > > [...] > > > > With this patch, we did boot a 16TiB machine. Without the patches, the > > v3.10 kernel with the same configuration took 407 seconds for > > free_all_bootmem. With the patches and operating on 2MiB pages instead > > of 1GiB, it took 26 seconds so performance was improved. I have no feel > > for how the 1GiB chunk size will perform. > > That's pretty impressive. And WRONG! That is a 15x speedup in the freeing of memory at the free_all_bootmem point. That is _NOT_ the speedup from memmap_init_zone. I forgot to take that into account as Nate pointed out this morning in a hallway discussion. Before, on the 16TiB machine, memmap_init_zone took 1152 seconds. After, it took 50. If it were a straight 1/512th, we would have expected that 1152 to be something more on the line of 2-3 so there is still significant room for improvement. Sorry for the confusion. > It's still a 15x speedup instead of a 512x speedup, so I'd say there's > something else being the current bottleneck, besides page init > granularity. > > Can you boot with just a few gigs of RAM and stuff the rest into hotplug > memory, and then hot-add that memory? That would allow easy profiling of > remaining overhead. Nate and I will be working on other things for the next few hours hoping there is a better answer to the first question we asked about there being a way to test a page other than comparing against all zeroes to see if it has been initialized. Thanks, Robin -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
next prev parent reply other threads:[~2013-07-15 15:16 UTC|newest] Thread overview: 153+ messages / expand[flat|nested] mbox.gz Atom feed top 2013-07-12 2:03 [RFC 0/4] Transparent on-demand struct page initialization embedded in the buddy allocator Robin Holt 2013-07-12 2:03 ` Robin Holt 2013-07-12 2:03 ` [RFC 1/4] memblock: Introduce a for_each_reserved_mem_region iterator Robin Holt 2013-07-12 2:03 ` Robin Holt 2013-07-12 2:03 ` [RFC 2/4] Have __free_pages_memory() free in larger chunks Robin Holt 2013-07-12 2:03 ` Robin Holt 2013-07-12 7:45 ` Robin Holt 2013-07-12 7:45 ` Robin Holt 2013-07-13 3:08 ` Yinghai Lu 2013-07-13 3:08 ` Yinghai Lu 2013-07-16 13:02 ` Sam Ben 2013-07-16 13:02 ` Sam Ben 2013-07-23 15:32 ` Johannes Weiner 2013-07-23 15:32 ` Johannes Weiner 2013-07-12 2:03 ` [RFC 3/4] Seperate page initialization into a separate function Robin Holt 2013-07-12 2:03 ` Robin Holt 2013-07-13 3:06 ` Yinghai Lu 2013-07-13 3:06 ` Yinghai Lu 2013-07-15 3:19 ` Robin Holt 2013-07-15 3:19 ` Robin Holt 2013-07-12 2:03 ` [RFC 4/4] Sparse initialization of struct page array Robin Holt 2013-07-12 2:03 ` Robin Holt 2013-07-13 4:19 ` Yinghai Lu 2013-07-13 4:19 ` Yinghai Lu 2013-07-13 4:39 ` H. Peter Anvin 2013-07-13 4:39 ` H. Peter Anvin 2013-07-13 5:31 ` Yinghai Lu 2013-07-13 5:31 ` Yinghai Lu 2013-07-13 5:38 ` H. Peter Anvin 2013-07-13 5:38 ` H. Peter Anvin 2013-07-15 14:08 ` Nathan Zimmer 2013-07-15 14:08 ` Nathan Zimmer 2013-07-15 17:45 ` Nathan Zimmer 2013-07-15 17:45 ` Nathan Zimmer 2013-07-15 17:54 ` H. Peter Anvin 2013-07-15 17:54 ` H. Peter Anvin 2013-07-15 18:26 ` Robin Holt 2013-07-15 18:26 ` Robin Holt 2013-07-15 18:29 ` H. Peter Anvin 2013-07-15 18:29 ` H. Peter Anvin 2013-07-23 8:32 ` Ingo Molnar 2013-07-23 8:32 ` Ingo Molnar 2013-07-23 11:09 ` Robin Holt 2013-07-23 11:09 ` Robin Holt 2013-07-23 11:15 ` Robin Holt 2013-07-23 11:15 ` Robin Holt 2013-07-23 11:41 ` Robin Holt 2013-07-23 11:41 ` Robin Holt 2013-07-23 11:50 ` Robin Holt 2013-07-23 11:50 ` Robin Holt 2013-07-16 10:26 ` Robin Holt 2013-07-16 10:26 ` Robin Holt 2013-07-25 2:25 ` Robin Holt 2013-07-25 2:25 ` Robin Holt 2013-07-25 12:50 ` Yinghai Lu 2013-07-25 12:50 ` Yinghai Lu 2013-07-25 13:42 ` Robin Holt 2013-07-25 13:42 ` Robin Holt 2013-07-25 13:52 ` Yinghai Lu 2013-07-25 13:52 ` Yinghai Lu 2013-07-15 21:30 ` Andrew Morton 2013-07-15 21:30 ` Andrew Morton 2013-07-16 10:38 ` Robin Holt 2013-07-16 10:38 ` Robin Holt 2013-07-12 8:27 ` [RFC 0/4] Transparent on-demand struct page initialization embedded in the buddy allocator Ingo Molnar 2013-07-12 8:27 ` Ingo Molnar 2013-07-12 8:47 ` boot tracing Borislav Petkov 2013-07-12 8:47 ` Borislav Petkov 2013-07-12 8:53 ` Ingo Molnar 2013-07-12 8:53 ` Ingo Molnar 2013-07-15 1:38 ` Sam Ben 2013-07-15 1:38 ` Sam Ben 2013-07-23 8:18 ` Ingo Molnar 2013-07-23 8:18 ` Ingo Molnar 2013-07-12 9:19 ` [RFC 0/4] Transparent on-demand struct page initialization embedded in the buddy allocator Robert Richter 2013-07-12 9:19 ` Robert Richter 2013-07-15 15:16 ` Robin Holt [this message] 2013-07-15 15:16 ` Robin Holt 2013-07-16 8:55 ` Joonsoo Kim 2013-07-16 8:55 ` Joonsoo Kim 2013-07-16 9:08 ` Borislav Petkov 2013-07-16 9:08 ` Borislav Petkov 2013-07-23 8:20 ` Ingo Molnar 2013-07-23 8:20 ` Ingo Molnar 2013-07-15 15:00 ` Robin Holt 2013-07-15 15:00 ` Robin Holt 2013-07-17 5:17 ` Sam Ben 2013-07-17 5:17 ` Sam Ben 2013-07-17 9:30 ` Robin Holt 2013-07-17 9:30 ` Robin Holt 2013-07-19 23:51 ` Yinghai Lu 2013-07-22 6:13 ` Robin Holt 2013-07-22 6:13 ` Robin Holt 2013-08-02 17:44 ` [RFC v2 0/5] " Nathan Zimmer 2013-08-02 17:44 ` Nathan Zimmer 2013-08-02 17:44 ` [RFC v2 1/5] memblock: Introduce a for_each_reserved_mem_region iterator Nathan Zimmer 2013-08-02 17:44 ` Nathan Zimmer 2013-08-02 17:44 ` [RFC v2 2/5] Have __free_pages_memory() free in larger chunks Nathan Zimmer 2013-08-02 17:44 ` Nathan Zimmer 2013-08-02 17:44 ` [RFC v2 3/5] Move page initialization into a separate function Nathan Zimmer 2013-08-02 17:44 ` Nathan Zimmer 2013-08-02 17:44 ` [RFC v2 4/5] Only set page reserved in the memblock region Nathan Zimmer 2013-08-02 17:44 ` Nathan Zimmer 2013-08-03 20:04 ` Nathan Zimmer 2013-08-03 20:04 ` Nathan Zimmer 2013-08-02 17:44 ` [RFC v2 5/5] Sparse initialization of struct page array Nathan Zimmer 2013-08-02 17:44 ` Nathan Zimmer 2013-08-05 9:58 ` [RFC v2 0/5] Transparent on-demand struct page initialization embedded in the buddy allocator Ingo Molnar 2013-08-05 9:58 ` Ingo Molnar 2013-08-12 21:54 ` [RFC v3 " Nathan Zimmer 2013-08-12 21:54 ` Nathan Zimmer 2013-08-12 21:54 ` [RFC v3 1/5] memblock: Introduce a for_each_reserved_mem_region iterator Nathan Zimmer 2013-08-12 21:54 ` Nathan Zimmer 2013-08-12 21:54 ` [RFC v3 2/5] Have __free_pages_memory() free in larger chunks Nathan Zimmer 2013-08-12 21:54 ` Nathan Zimmer 2013-08-12 21:54 ` [RFC v3 3/5] Move page initialization into a separate function Nathan Zimmer 2013-08-12 21:54 ` Nathan Zimmer 2013-08-12 21:54 ` [RFC v3 4/5] Only set page reserved in the memblock region Nathan Zimmer 2013-08-12 21:54 ` Nathan Zimmer 2013-08-12 21:54 ` [RFC v3 5/5] Sparse initialization of struct page array Nathan Zimmer 2013-08-12 21:54 ` Nathan Zimmer 2013-08-13 10:58 ` [RFC v3 0/5] Transparent on-demand struct page initialization embedded in the buddy allocator Ingo Molnar 2013-08-13 10:58 ` Ingo Molnar 2013-08-13 17:09 ` Linus Torvalds 2013-08-13 17:09 ` Linus Torvalds 2013-08-13 17:23 ` H. Peter Anvin 2013-08-13 17:23 ` H. Peter Anvin 2013-08-13 17:33 ` Mike Travis 2013-08-13 17:33 ` Mike Travis 2013-08-13 17:51 ` Linus Torvalds 2013-08-13 17:51 ` Linus Torvalds 2013-08-13 18:04 ` Mike Travis 2013-08-13 18:04 ` Mike Travis 2013-08-13 19:06 ` Mike Travis 2013-08-13 19:06 ` Mike Travis 2013-08-13 20:24 ` Yinghai Lu 2013-08-13 20:24 ` Yinghai Lu 2013-08-13 20:37 ` Mike Travis 2013-08-13 20:37 ` Mike Travis 2013-08-13 21:35 ` Nathan Zimmer 2013-08-13 21:35 ` Nathan Zimmer 2013-08-13 23:10 ` Nathan Zimmer 2013-08-13 23:10 ` Nathan Zimmer 2013-08-13 23:55 ` Linus Torvalds 2013-08-13 23:55 ` Linus Torvalds 2013-08-14 11:27 ` Ingo Molnar 2013-08-14 11:27 ` Ingo Molnar 2013-08-14 11:05 ` Ingo Molnar 2013-08-14 11:05 ` Ingo Molnar 2013-08-14 22:15 ` Nathan Zimmer 2013-08-14 22:15 ` Nathan Zimmer 2013-08-16 16:36 ` Dave Hansen 2013-08-16 16:36 ` Dave Hansen
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=20130715151623.GB3421@sgi.com \ --to=holt@sgi.com \ --cc=a.p.zijlstra@chello.nl \ --cc=akpm@linux-foundation.org \ --cc=bp@alien8.de \ --cc=daniel@numascale-asia.com \ --cc=gregkh@linuxfoundation.org \ --cc=hpa@zytor.com \ --cc=linux-kernel@vger.kernel.org \ --cc=linux-mm@kvack.org \ --cc=mgorman@suse.de \ --cc=mingo@kernel.org \ --cc=nzimmer@sgi.com \ --cc=rob@landley.net \ --cc=rric@kernel.org \ --cc=travis@sgi.com \ --cc=yinghai@kernel.org \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: linkBe sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.