| Message ID | 20190610111252.239156-1-minchan@kernel.org (mailing list archive) |
|---|---|
| Headers | show |
| Series | Introduce MADV_COLD and MADV_PAGEOUT | expand |
I'd really love to see the manpages for these new flags. The devil is in the details of our promises to userspace.
Hi! > - Problem > > Naturally, cached apps were dominant consumers of memory on the system. > However, they were not significant consumers of swap even though they are > good candidate for swap. Under investigation, swapping out only begins > once the low zone watermark is hit and kswapd wakes up, but the overall > allocation rate in the system might trip lmkd thresholds and cause a cached > process to be killed(we measured performance swapping out vs. zapping the > memory by killing a process. Unsurprisingly, zapping is 10x times faster > even though we use zram which is much faster than real storage) so kill > from lmkd will often satisfy the high zone watermark, resulting in very > few pages actually being moved to swap. Is it still faster to swap-in the application than to restart it? > This approach is similar in spirit to madvise(MADV_WONTNEED), but the > information required to make the reclaim decision is not known to the app. > Instead, it is known to a centralized userspace daemon, and that daemon > must be able to initiate reclaim on its own without any app involvement. > To solve the concern, this patch introduces new syscall - > > struct pr_madvise_param { > int size; /* the size of this structure */ > int cookie; /* reserved to support atomicity */ > int nr_elem; /* count of below arrary fields */ > int __user *hints; /* hints for each range */ > /* to store result of each operation */ > const struct iovec __user *results; > /* input address ranges */ > const struct iovec __user *ranges; > }; > > int process_madvise(int pidfd, struct pr_madvise_param *u_param, > unsigned long flags); That's quite a complex interface. Could we simply have feel_free_to_swap_out(int pid) syscall? :-). Pavel
On Wed, Jun 12, 2019 at 12:59:45PM +0200, Pavel Machek wrote: > > - Problem > > > > Naturally, cached apps were dominant consumers of memory on the system. > > However, they were not significant consumers of swap even though they are > > good candidate for swap. Under investigation, swapping out only begins > > once the low zone watermark is hit and kswapd wakes up, but the overall > > allocation rate in the system might trip lmkd thresholds and cause a cached > > process to be killed(we measured performance swapping out vs. zapping the > > memory by killing a process. Unsurprisingly, zapping is 10x times faster > > even though we use zram which is much faster than real storage) so kill > > from lmkd will often satisfy the high zone watermark, resulting in very > > few pages actually being moved to swap. > > Is it still faster to swap-in the application than to restart it? It's the same type of question I was addressing earlier in the remote KSM discussion: making applications aware of all the memory management stuff or delegate the decision to some supervising task. In this case, we cannot rewrite all the application to handle imaginary SIGRESTART (or whatever you invent to handle restarts gracefully). SIGTERM may require more memory to finish stuff to not lose your data (and I guess you don't want to lose your data, right?), and SIGKILL is pretty much destructive. Offloading proactive memory management to a process that knows how to do it allows to handle not only throwaway containers/microservices, but also usual desktop/mobile workflow. > > This approach is similar in spirit to madvise(MADV_WONTNEED), but the > > information required to make the reclaim decision is not known to the app. > > Instead, it is known to a centralized userspace daemon, and that daemon > > must be able to initiate reclaim on its own without any app involvement. > > To solve the concern, this patch introduces new syscall - > > > > struct pr_madvise_param { > > int size; /* the size of this structure */ > > int cookie; /* reserved to support atomicity */ > > int nr_elem; /* count of below arrary fields */ > > int __user *hints; /* hints for each range */ > > /* to store result of each operation */ > > const struct iovec __user *results; > > /* input address ranges */ > > const struct iovec __user *ranges; > > }; > > > > int process_madvise(int pidfd, struct pr_madvise_param *u_param, > > unsigned long flags); > > That's quite a complex interface. > > Could we simply have feel_free_to_swap_out(int pid) syscall? :-). I wonder for how long we'll go on with adding new syscalls each time we need some amendment to existing interfaces. Yes, clone6(), I'm looking at you :(. In case of process_madvise() keep in mind it will be focused not only on MADV_COLD, but also, potentially, on other MADV_ flags as well. I can hardly imagine we'll add one syscall per each flag.
Hi! > > > This approach is similar in spirit to madvise(MADV_WONTNEED), but the > > > information required to make the reclaim decision is not known to the app. > > > Instead, it is known to a centralized userspace daemon, and that daemon > > > must be able to initiate reclaim on its own without any app involvement. > > > To solve the concern, this patch introduces new syscall - > > > > > > struct pr_madvise_param { > > > int size; /* the size of this structure */ > > > int cookie; /* reserved to support atomicity */ > > > int nr_elem; /* count of below arrary fields */ > > > int __user *hints; /* hints for each range */ > > > /* to store result of each operation */ > > > const struct iovec __user *results; > > > /* input address ranges */ > > > const struct iovec __user *ranges; > > > }; > > > > > > int process_madvise(int pidfd, struct pr_madvise_param *u_param, > > > unsigned long flags); > > > > That's quite a complex interface. > > > > Could we simply have feel_free_to_swap_out(int pid) syscall? :-). > > I wonder for how long we'll go on with adding new syscalls each time we need > some amendment to existing interfaces. Yes, clone6(), I'm looking at > you :(. > > In case of process_madvise() keep in mind it will be focused not only on > MADV_COLD, but also, potentially, on other MADV_ flags as well. I can > hardly imagine we'll add one syscall per each flag. Use case described above talked about whole-process-at-a-time usage, so I'm asking if simpler interface/code is enough. If there's motivation for more complex version, it should be described here... Pavel
On Mon, Jun 10, 2019 at 11:03:00AM -0700, Dave Hansen wrote: > I'd really love to see the manpages for these new flags. The devil is > in the details of our promises to userspace. I'm waiting comments from reviewers since I have fixed what they point out from the previous version. I will add manpage material in respin after the getting more feedback. Thanks.
On Mon 10-06-19 20:12:47, Minchan Kim wrote: > This patch is part of previous series: > https://lore.kernel.org/lkml/20190531064313.193437-1-minchan@kernel.org/T/#u > Originally, it was created for external madvise hinting feature. > > https://lkml.org/lkml/2019/5/31/463 > Michal wanted to separte the discussion from external hinting interface > so this patchset includes only first part of my entire patchset > > - introduce MADV_COLD and MADV_PAGEOUT hint to madvise. > > However, I keep entire description for others for easier understanding > why this kinds of hint was born. > > Thanks. > > This patchset is against on next-20190530. > > Below is description of previous entire patchset. > ================= &< ===================== > > - Background > > The Android terminology used for forking a new process and starting an app > from scratch is a cold start, while resuming an existing app is a hot start. > While we continually try to improve the performance of cold starts, hot > starts will always be significantly less power hungry as well as faster so > we are trying to make hot start more likely than cold start. > > To increase hot start, Android userspace manages the order that apps should > be killed in a process called ActivityManagerService. ActivityManagerService > tracks every Android app or service that the user could be interacting with > at any time and translates that into a ranked list for lmkd(low memory > killer daemon). They are likely to be killed by lmkd if the system has to > reclaim memory. In that sense they are similar to entries in any other cache. > Those apps are kept alive for opportunistic performance improvements but > those performance improvements will vary based on the memory requirements of > individual workloads. > > - Problem > > Naturally, cached apps were dominant consumers of memory on the system. > However, they were not significant consumers of swap even though they are > good candidate for swap. Under investigation, swapping out only begins > once the low zone watermark is hit and kswapd wakes up, but the overall > allocation rate in the system might trip lmkd thresholds and cause a cached > process to be killed(we measured performance swapping out vs. zapping the > memory by killing a process. Unsurprisingly, zapping is 10x times faster > even though we use zram which is much faster than real storage) so kill > from lmkd will often satisfy the high zone watermark, resulting in very > few pages actually being moved to swap. > > - Approach > > The approach we chose was to use a new interface to allow userspace to > proactively reclaim entire processes by leveraging platform information. > This allowed us to bypass the inaccuracy of the kernel’s LRUs for pages > that are known to be cold from userspace and to avoid races with lmkd > by reclaiming apps as soon as they entered the cached state. Additionally, > it could provide many chances for platform to use much information to > optimize memory efficiency. > > To achieve the goal, the patchset introduce two new options for madvise. > One is MADV_COLD which will deactivate activated pages and the other is > MADV_PAGEOUT which will reclaim private pages instantly. These new options > complement MADV_DONTNEED and MADV_FREE by adding non-destructive ways to > gain some free memory space. MADV_PAGEOUT is similar to MADV_DONTNEED in a way > that it hints the kernel that memory region is not currently needed and > should be reclaimed immediately; MADV_COLD is similar to MADV_FREE in a way > that it hints the kernel that memory region is not currently needed and > should be reclaimed when memory pressure rises. This all is a very good background information suitable for the cover letter. > This approach is similar in spirit to madvise(MADV_WONTNEED), but the > information required to make the reclaim decision is not known to the app. > Instead, it is known to a centralized userspace daemon, and that daemon > must be able to initiate reclaim on its own without any app involvement. > To solve the concern, this patch introduces new syscall - > > struct pr_madvise_param { > int size; /* the size of this structure */ > int cookie; /* reserved to support atomicity */ > int nr_elem; /* count of below arrary fields */ > int __user *hints; /* hints for each range */ > /* to store result of each operation */ > const struct iovec __user *results; > /* input address ranges */ > const struct iovec __user *ranges; > }; > > int process_madvise(int pidfd, struct pr_madvise_param *u_param, > unsigned long flags); But this and the following paragraphs are referring to the later step when the madvise gains a remote process capabilities and that is out of the scope of this patch series so I would simply remove it from here. Andrew tends to put the cover letter into the first patch of the series and that would be indeed confusing here.
On Wed, Jun 19, 2019 at 02:27:50PM +0200, Michal Hocko wrote: > On Mon 10-06-19 20:12:47, Minchan Kim wrote: > > This patch is part of previous series: > > https://lore.kernel.org/lkml/20190531064313.193437-1-minchan@kernel.org/T/#u > > Originally, it was created for external madvise hinting feature. > > > > https://lkml.org/lkml/2019/5/31/463 > > Michal wanted to separte the discussion from external hinting interface > > so this patchset includes only first part of my entire patchset > > > > - introduce MADV_COLD and MADV_PAGEOUT hint to madvise. > > > > However, I keep entire description for others for easier understanding > > why this kinds of hint was born. > > > > Thanks. > > > > This patchset is against on next-20190530. > > > > Below is description of previous entire patchset. > > ================= &< ===================== > > > > - Background > > > > The Android terminology used for forking a new process and starting an app > > from scratch is a cold start, while resuming an existing app is a hot start. > > While we continually try to improve the performance of cold starts, hot > > starts will always be significantly less power hungry as well as faster so > > we are trying to make hot start more likely than cold start. > > > > To increase hot start, Android userspace manages the order that apps should > > be killed in a process called ActivityManagerService. ActivityManagerService > > tracks every Android app or service that the user could be interacting with > > at any time and translates that into a ranked list for lmkd(low memory > > killer daemon). They are likely to be killed by lmkd if the system has to > > reclaim memory. In that sense they are similar to entries in any other cache. > > Those apps are kept alive for opportunistic performance improvements but > > those performance improvements will vary based on the memory requirements of > > individual workloads. > > > > - Problem > > > > Naturally, cached apps were dominant consumers of memory on the system. > > However, they were not significant consumers of swap even though they are > > good candidate for swap. Under investigation, swapping out only begins > > once the low zone watermark is hit and kswapd wakes up, but the overall > > allocation rate in the system might trip lmkd thresholds and cause a cached > > process to be killed(we measured performance swapping out vs. zapping the > > memory by killing a process. Unsurprisingly, zapping is 10x times faster > > even though we use zram which is much faster than real storage) so kill > > from lmkd will often satisfy the high zone watermark, resulting in very > > few pages actually being moved to swap. > > > > - Approach > > > > The approach we chose was to use a new interface to allow userspace to > > proactively reclaim entire processes by leveraging platform information. > > This allowed us to bypass the inaccuracy of the kernel’s LRUs for pages > > that are known to be cold from userspace and to avoid races with lmkd > > by reclaiming apps as soon as they entered the cached state. Additionally, > > it could provide many chances for platform to use much information to > > optimize memory efficiency. > > > > To achieve the goal, the patchset introduce two new options for madvise. > > One is MADV_COLD which will deactivate activated pages and the other is > > MADV_PAGEOUT which will reclaim private pages instantly. These new options > > complement MADV_DONTNEED and MADV_FREE by adding non-destructive ways to > > gain some free memory space. MADV_PAGEOUT is similar to MADV_DONTNEED in a way > > that it hints the kernel that memory region is not currently needed and > > should be reclaimed immediately; MADV_COLD is similar to MADV_FREE in a way > > that it hints the kernel that memory region is not currently needed and > > should be reclaimed when memory pressure rises. > > This all is a very good background information suitable for the cover > letter. > > > This approach is similar in spirit to madvise(MADV_WONTNEED), but the > > information required to make the reclaim decision is not known to the app. > > Instead, it is known to a centralized userspace daemon, and that daemon > > must be able to initiate reclaim on its own without any app involvement. > > To solve the concern, this patch introduces new syscall - > > > > struct pr_madvise_param { > > int size; /* the size of this structure */ > > int cookie; /* reserved to support atomicity */ > > int nr_elem; /* count of below arrary fields */ > > int __user *hints; /* hints for each range */ > > /* to store result of each operation */ > > const struct iovec __user *results; > > /* input address ranges */ > > const struct iovec __user *ranges; > > }; > > > > int process_madvise(int pidfd, struct pr_madvise_param *u_param, > > unsigned long flags); > > But this and the following paragraphs are referring to the later step > when the madvise gains a remote process capabilities and that is out > of the scope of this patch series so I would simply remove it from > here. Andrew tends to put the cover letter into the first patch of the > series and that would be indeed > confusing here. Okay, I will remove the part in next revision.
This patch is part of previous series: https://lore.kernel.org/lkml/20190531064313.193437-1-minchan@kernel.org/T/#u Originally, it was created for external madvise hinting feature. https://lkml.org/lkml/2019/5/31/463 Michal wanted to separte the discussion from external hinting interface so this patchset includes only first part of my entire patchset - introduce MADV_COLD and MADV_PAGEOUT hint to madvise. However, I keep entire description for others for easier understanding why this kinds of hint was born. Thanks. This patchset is against on next-20190530. Below is description of previous entire patchset. ================= &< ===================== - Background The Android terminology used for forking a new process and starting an app from scratch is a cold start, while resuming an existing app is a hot start. While we continually try to improve the performance of cold starts, hot starts will always be significantly less power hungry as well as faster so we are trying to make hot start more likely than cold start. To increase hot start, Android userspace manages the order that apps should be killed in a process called ActivityManagerService. ActivityManagerService tracks every Android app or service that the user could be interacting with at any time and translates that into a ranked list for lmkd(low memory killer daemon). They are likely to be killed by lmkd if the system has to reclaim memory. In that sense they are similar to entries in any other cache. Those apps are kept alive for opportunistic performance improvements but those performance improvements will vary based on the memory requirements of individual workloads. - Problem Naturally, cached apps were dominant consumers of memory on the system. However, they were not significant consumers of swap even though they are good candidate for swap. Under investigation, swapping out only begins once the low zone watermark is hit and kswapd wakes up, but the overall allocation rate in the system might trip lmkd thresholds and cause a cached process to be killed(we measured performance swapping out vs. zapping the memory by killing a process. Unsurprisingly, zapping is 10x times faster even though we use zram which is much faster than real storage) so kill from lmkd will often satisfy the high zone watermark, resulting in very few pages actually being moved to swap. - Approach The approach we chose was to use a new interface to allow userspace to proactively reclaim entire processes by leveraging platform information. This allowed us to bypass the inaccuracy of the kernel’s LRUs for pages that are known to be cold from userspace and to avoid races with lmkd by reclaiming apps as soon as they entered the cached state. Additionally, it could provide many chances for platform to use much information to optimize memory efficiency. To achieve the goal, the patchset introduce two new options for madvise. One is MADV_COLD which will deactivate activated pages and the other is MADV_PAGEOUT which will reclaim private pages instantly. These new options complement MADV_DONTNEED and MADV_FREE by adding non-destructive ways to gain some free memory space. MADV_PAGEOUT is similar to MADV_DONTNEED in a way that it hints the kernel that memory region is not currently needed and should be reclaimed immediately; MADV_COLD is similar to MADV_FREE in a way that it hints the kernel that memory region is not currently needed and should be reclaimed when memory pressure rises. This approach is similar in spirit to madvise(MADV_WONTNEED), but the information required to make the reclaim decision is not known to the app. Instead, it is known to a centralized userspace daemon, and that daemon must be able to initiate reclaim on its own without any app involvement. To solve the concern, this patch introduces new syscall - struct pr_madvise_param { int size; /* the size of this structure */ int cookie; /* reserved to support atomicity */ int nr_elem; /* count of below arrary fields */ int __user *hints; /* hints for each range */ /* to store result of each operation */ const struct iovec __user *results; /* input address ranges */ const struct iovec __user *ranges; }; int process_madvise(int pidfd, struct pr_madvise_param *u_param, unsigned long flags); The syscall get pidfd to give hints to external process and provides pair of result/ranges vector arguments so that it could give several hints to each address range all at once. It also has cookie variable to support atomicity of the API for address ranges operations. IOW, if target process changes address space since monitor process has parsed address ranges via map_files or maps, the API can detect the race so could cancel entire address space operation. It's not implemented yet. Daniel Colascione suggested a idea(Please read description in patch[6/6]) and this patchset adds cookie a variable for the future. - Experiment We did bunch of testing with several hundreds of real users, not artificial benchmark on android. We saw about 17% cold start decreasement without any significant battery/app startup latency issues. And with artificial benchmark which launches and switching apps, we saw average 7% app launching improvement, 18% less lmkd kill and good stat from vmstat. A is vanilla and B is process_madvise. A B delta ratio(%) allocstall_dma 0 0 0 0.00 allocstall_movable 1464 457 -1007 -69.00 allocstall_normal 263210 190763 -72447 -28.00 allocstall_total 264674 191220 -73454 -28.00 compact_daemon_wake 26912 25294 -1618 -7.00 compact_fail 17885 14151 -3734 -21.00 compact_free_scanned 4204766409 3835994922 -368771487 -9.00 compact_isolated 3446484 2967618 -478866 -14.00 compact_migrate_scanned 1621336411 1324695710 -296640701 -19.00 compact_stall 19387 15343 -4044 -21.00 compact_success 1502 1192 -310 -21.00 kswapd_high_wmark_hit_quickly 234 184 -50 -22.00 kswapd_inodesteal 221635 233093 11458 5.00 kswapd_low_wmark_hit_quickly 66065 54009 -12056 -19.00 nr_dirtied 259934 296476 36542 14.00 nr_vmscan_immediate_reclaim 2587 2356 -231 -9.00 nr_vmscan_write 1274232 2661733 1387501 108.00 nr_written 1514060 2937560 1423500 94.00 pageoutrun 67561 55133 -12428 -19.00 pgactivate 2335060 1984882 -350178 -15.00 pgalloc_dma 13743011 14096463 353452 2.00 pgalloc_movable 0 0 0 0.00 pgalloc_normal 18742440 16802065 -1940375 -11.00 pgalloc_total 32485451 30898528 -1586923 -5.00 pgdeactivate 4262210 2930670 -1331540 -32.00 pgfault 30812334 31085065 272731 0.00 pgfree 33553970 31765164 -1788806 -6.00 pginodesteal 33411 15084 -18327 -55.00 pglazyfreed 0 0 0 0.00 pgmajfault 551312 1508299 956987 173.00 pgmigrate_fail 43927 29330 -14597 -34.00 pgmigrate_success 1399851 1203922 -195929 -14.00 pgpgin 24141776 19032156 -5109620 -22.00 pgpgout 959344 1103316 143972 15.00 pgpgoutclean 4639732 3765868 -873864 -19.00 pgrefill 4884560 3006938 -1877622 -39.00 pgrotated 37828 25897 -11931 -32.00 pgscan_direct 1456037 957567 -498470 -35.00 pgscan_direct_throttle 0 0 0 0.00 pgscan_kswapd 6667767 5047360 -1620407 -25.00 pgscan_total 8123804 6004927 -2118877 -27.00 pgskip_dma 0 0 0 0.00 pgskip_movable 0 0 0 0.00 pgskip_normal 14907 25382 10475 70.00 pgskip_total 14907 25382 10475 70.00 pgsteal_direct 1118986 690215 -428771 -39.00 pgsteal_kswapd 4750223 3657107 -1093116 -24.00 pgsteal_total 5869209 4347322 -1521887 -26.00 pswpin 417613 1392647 975034 233.00 pswpout 1274224 2661731 1387507 108.00 slabs_scanned 13686905 10807200 -2879705 -22.00 workingset_activate 668966 569444 -99522 -15.00 workingset_nodereclaim 38957 32621 -6336 -17.00 workingset_refault 2816795 2179782 -637013 -23.00 workingset_restore 294320 168601 -125719 -43.00 pgmajfault is increased by 173% because swapin is increased by 200% by process_madvise hint. However, swap read based on zram is much cheaper than file IO in performance point of view and app hot start by swapin is also cheaper than cold start from the beginning of app which needs many IO from storage and initialization steps. Brian Geffon in ChromeOS team had an experiment with process_madvise(2) Quote form him: "What I found is that by using process_madvise after a tab has been back grounded for more than 45 seconds reduced the average tab switch times by 25%! This is a huge result and very obvious validation that process_madvise hints works well for the ChromeOS use case." This patchset is against on next-20190607. Minchan Kim (5): mm: introduce MADV_COLD mm: change PAGEREF_RECLAIM_CLEAN with PAGE_REFRECLAIM mm: account nr_isolated_xxx in [isolate|putback]_lru_page mm: introduce MADV_PAGEOUT mm: factor out pmd young/dirty bit handling and THP split include/linux/huge_mm.h | 3 - include/linux/swap.h | 2 + include/uapi/asm-generic/mman-common.h | 2 + mm/compaction.c | 2 - mm/gup.c | 7 +- mm/huge_memory.c | 74 ----- mm/internal.h | 2 +- mm/khugepaged.c | 3 - mm/madvise.c | 358 ++++++++++++++++++++++++- mm/memory-failure.c | 3 - mm/memory_hotplug.c | 4 - mm/mempolicy.c | 6 +- mm/migrate.c | 37 +-- mm/oom_kill.c | 2 +- mm/swap.c | 42 +++ mm/vmscan.c | 86 +++++- 16 files changed, 486 insertions(+), 147 deletions(-)