From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753953AbcITDGA (ORCPT ); Mon, 19 Sep 2016 23:06:00 -0400 Received: from szxga01-in.huawei.com ([58.251.152.64]:57852 "EHLO szxga01-in.huawei.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753816AbcITDF6 (ORCPT ); Mon, 19 Sep 2016 23:05:58 -0400 Subject: Re: [PATCH] f2fs: fix to avoid slowing down background gc To: Jaegeuk Kim References: <20160918115227.130658-1-yuchao0@huawei.com> <20160919221211.GC68913@jaegeuk> <20160920025432.GB72323@jaegeuk> CC: , From: Chao Yu Message-ID: <657c16a3-a63a-476a-d0a4-440c5c17c625@huawei.com> Date: Tue, 20 Sep 2016 11:05:46 +0800 User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:45.0) Gecko/20100101 Thunderbird/45.0 MIME-Version: 1.0 In-Reply-To: <20160920025432.GB72323@jaegeuk> Content-Type: text/plain; charset="windows-1252" Content-Transfer-Encoding: 7bit X-Originating-IP: [10.134.22.195] X-CFilter-Loop: Reflected X-Mirapoint-Virus-RAPID-Raw: score=unknown(0), refid=str=0001.0A020201.57E0A790.00B7,ss=1,re=0.000,recu=0.000,reip=0.000,cl=1,cld=1,fgs=0, ip=0.0.0.0, so=2013-06-18 04:22:30, dmn=2013-03-21 17:37:32 X-Mirapoint-Loop-Id: fed5ac758d2728215497bfa966c9e2fc Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On 2016/9/20 10:54, Jaegeuk Kim wrote: > On Tue, Sep 20, 2016 at 10:22:22AM +0800, Chao Yu wrote: >> Hi Jaegeuk, >> >> On 2016/9/20 6:12, Jaegeuk Kim wrote: >>> Hi Chao, >>> >>> On Sun, Sep 18, 2016 at 07:52:27PM +0800, Chao Yu wrote: >>>> Previously, we will choose to speed up background gc when the below >>>> conditions are both satisfied: >>>> a. There are a number of invalid blocks >>>> b. There is not enough free space >>>> >>>> But, when space utilization is high (utilization > 60%), there will be >>>> not enough invalid blocks, result in slowing down background gc, after >>>> then there are more opportunities that triggering foreground gc due to >>>> high fragmented free space in fs. >>>> >>>> Remove condition a) in order to avoid slow down background gc speed in >>>> a high utilization fs. >>> >>> There exists a trade-off here: wear-out vs. eager gc for future speed-up. >>> How about using a kind of f2fs's dirty level (e.g., BDF)? >> >> Yep, I think that f2fs can implement a mechanism which can provide more >> dynamically adjustable GC speed in the specified scenario of user, by this, user >> can choose the strategy which is more beneficial to aspect >> (wear-out/performance) they care. Let me think a while, anyway I agree that BDF >> is a good reference value here. >> >> And Before we can provide above ability, how about treat this patch as a fixing >> patch, since it fixes to not adjust speed of GC according to utilization watermark? > > Well, this is not a bug fix, but a very conservative policy. So, please let's > make a better policy, if possible. Alright, let's think about this. Thanks, > > Thanks, > >> >> Thanks, >> >>> >>> Thanks, >>> >>>> >>>> Signed-off-by: Chao Yu >>>> --- >>>> fs/f2fs/gc.h | 18 +++--------------- >>>> 1 file changed, 3 insertions(+), 15 deletions(-) >>>> >>>> diff --git a/fs/f2fs/gc.h b/fs/f2fs/gc.h >>>> index a993967..5d0a19c 100644 >>>> --- a/fs/f2fs/gc.h >>>> +++ b/fs/f2fs/gc.h >>>> @@ -16,7 +16,6 @@ >>>> #define DEF_GC_THREAD_MIN_SLEEP_TIME 30000 /* milliseconds */ >>>> #define DEF_GC_THREAD_MAX_SLEEP_TIME 60000 >>>> #define DEF_GC_THREAD_NOGC_SLEEP_TIME 300000 /* wait 5 min */ >>>> -#define LIMIT_INVALID_BLOCK 40 /* percentage over total user space */ >>>> #define LIMIT_FREE_BLOCK 40 /* percentage over invalid + free space */ >>>> >>>> /* Search max. number of dirty segments to select a victim segment */ >>>> @@ -52,11 +51,6 @@ static inline block_t free_user_blocks(struct f2fs_sb_info *sbi) >>>> << sbi->log_blocks_per_seg; >>>> } >>>> >>>> -static inline block_t limit_invalid_user_blocks(struct f2fs_sb_info *sbi) >>>> -{ >>>> - return (long)(sbi->user_block_count * LIMIT_INVALID_BLOCK) / 100; >>>> -} >>>> - >>>> static inline block_t limit_free_user_blocks(struct f2fs_sb_info *sbi) >>>> { >>>> block_t reclaimable_user_blocks = sbi->user_block_count - >>>> @@ -88,15 +82,9 @@ static inline void decrease_sleep_time(struct f2fs_gc_kthread *gc_th, >>>> >>>> static inline bool has_enough_invalid_blocks(struct f2fs_sb_info *sbi) >>>> { >>>> - block_t invalid_user_blocks = sbi->user_block_count - >>>> - written_block_count(sbi); >>>> /* >>>> - * Background GC is triggered with the following conditions. >>>> - * 1. There are a number of invalid blocks. >>>> - * 2. There is not enough free space. >>>> + * Background GC should speed up when there is not enough free blocks >>>> + * in total unused (free + invalid) blocks. >>>> */ >>>> - if (invalid_user_blocks > limit_invalid_user_blocks(sbi) && >>>> - free_user_blocks(sbi) < limit_free_user_blocks(sbi)) >>>> - return true; >>>> - return false; >>>> + return free_user_blocks(sbi) < limit_free_user_blocks(sbi); >>>> } >>>> -- >>>> 2.8.2.311.gee88674 >>> >>> . >>> > > . >