From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jiro SEKIBA Subject: Re: [PATCH] nilfs2: tune write performance Date: Sun, 09 Aug 2009 23:04:13 +0900 Message-ID: <874oshdqxu.wl%jir@sekiba.com> References: <1249483023-21088-1-git-send-email-jir@unicus.jp> <20090806.035405.08300632.ryusuke@osrg.net> Reply-To: NILFS Users mailing list Mime-Version: 1.0 (generated by SEMI 1.14.6 - "Maruoka") Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20090806.035405.08300632.ryusuke-sG5X7nlA6pw@public.gmane.org> List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: users-bounces-JrjvKiOkagjYtjvyW6yDsg@public.gmane.org Errors-To: users-bounces-JrjvKiOkagjYtjvyW6yDsg@public.gmane.org To: NILFS Users mailing list Hi, Thank you for the comments! I'll update the patch and re-do the performance benchmark. At Thu, 06 Aug 2009 03:54:05 +0900 (JST), Ryusuke Konishi wrote: > > Hi! > On Wed, 5 Aug 2009 23:37:03 +0900, Jiro SEKIBA wrote: > > Hi, > > > > This is a candidate patch to improve write performance. > > > > When GC is runnning, GC moves live block to difference segments. > > Moving live blocks to different segment is done in a transaction, > > but it is not necessarily to be in the transaction. > > This patch will get the nilfs_ioctl_move_blocks() out from > > transaction lock and put it before the transaction. > > > > Here are the bonnie++ bench mark results against rc5. > > I used sata disk so performace difference is not big. > > So I modified nilfs_cleanerd.conf to ensure lock will be long enough. > > > > I ran "bonnie++ -b" three times in each environment. > > Create clean nilfs2 partition each time before test run. > > > > Results > > - write performance of rc5 > > * Char|17472.333 K/sec > > * Block|30748.000 K/sec > > * Rewrite|18506.000 K/sec > > > > - write performance of the patched rc5 > > * Char|17961.000 K/sec -> 102.700% > > * Block|30987.666 K/sec -> 100.700% > > * Rewrite|18775.666 K/sec -> 101.400% > > Thank you for the patch. > > I think this patch can mitigate freeze-time of writes due to GC, but > this benchmark seems not to be suited for measuring the effect. > Particularly, GC fires only once per five seconds with the following > config, and this may limit the efficacy in this kind of integrative > test. > > Is it possible to compare single write time hit GC with the time not > hit GC ? > > I will check correctness of the patch. > > Thanks, > Ryusuke Konishi > > > Here are the nilfs_cleanerd.conf and raw data. > > -------8<---------8<---------nilfs_cleanerd.conf---------8<---------8<------- > > protection_period 150 > > selection_policy timestamp # timestamp in ascend order > > nsegments_per_clean 10 > > cleaning_interval 5 > > retry_interval 60 > > use_mmap > > log_priority info > > -------8<---------8<---------nilfs_cleanerd.conf---------8<---------8<------- > > ---------8<---------8<---------rc5---------8<---------8<--------- > > Version 1.03d ------Sequential Output------ --Sequential Input- --Random- > > -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks-- > > Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP > > foras 10G 17404 22 30751 5 18519 3 33855 38 45560 4 47.3 0 > > ------Sequential Create------ --------Random Create-------- > > -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete-- > > files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP > > 16 103 1 +++++ +++ 255 0 251 4 +++++ +++ 223 1 > > foras,10G,17404,22,30751,5,18519,3,33855,38,45560,4,47.3,0,16,103,1,+++++,+++,255,0,251,4,+++++,+++,223,1 > > Version 1.03d ------Sequential Output------ --Sequential Input- --Random- > > -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks-- > > Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP > > foras 10G 17561 22 30781 5 18412 3 33973 38 46063 3 46.6 0 > > ------Sequential Create------ --------Random Create-------- > > -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete-- > > files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP > > 16 102 1 +++++ +++ 237 0 244 3 +++++ +++ 222 1 > > foras,10G,17561,22,30781,5,18412,3,33973,38,46063,3,46.6,0,16,102,1,+++++,+++,237,0,244,3,+++++,+++,222,1 > > Version 1.03d ------Sequential Output------ --Sequential Input- --Random- > > -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks-- > > Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP > > foras 10G 17452 22 30712 5 18587 3 34071 38 45479 4 50.6 0 > > ------Sequential Create------ --------Random Create-------- > > -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete-- > > files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP > > 16 102 1 +++++ +++ 238 0 252 4 +++++ +++ 222 1 > > foras,10G,17452,22,30712,5,18587,3,34071,38,45479,4,50.6,0,16,102,1,+++++,+++,238,0,252,4,+++++,+++,222,1 > > ---------8<---------8<---------rc5---------8<---------8<--------- > > ---------8<---------8<---------patch---------8<---------8<--------- > > Version 1.03d ------Sequential Output------ --Sequential Input- --Random- > > -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks-- > > Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP > > foras 10G 18004 22 31375 5 19110 3 33755 38 45224 4 97.1 0 > > ------Sequential Create------ --------Random Create-------- > > -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete-- > > files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP > > 16 99 1 +++++ +++ 223 0 231 3 +++++ +++ 209 1 > > foras,10G,18004,22,31375,5,19110,3,33755,38,45224,4,97.1,0,16,99,1,+++++,+++,223,0,231,3,+++++,+++,209,1 > > Version 1.03d ------Sequential Output------ --Sequential Input- --Random- > > -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks-- > > Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP > > foras 10G 17866 22 30673 5 18619 3 33923 38 45878 4 55.3 0 > > ------Sequential Create------ --------Random Create-------- > > -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete-- > > files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP > > 16 97 1 +++++ +++ 211 0 224 3 +++++ +++ 203 1 > > foras,10G,17866,22,30673,5,18619,3,33923,38,45878,4,55.3,0,16,97,1,+++++,+++,211,0,224,3,+++++,+++,203,1 > > Version 1.03d ------Sequential Output------ --Sequential Input- --Random- > > -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks-- > > Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP > > foras 10G 18013 22 30915 5 18598 3 33879 38 44957 4 47.5 0 > > ------Sequential Create------ --------Random Create-------- > > -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete-- > > files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP > > 16 99 1 +++++ +++ 255 0 253 4 +++++ +++ 226 1 > > foras,10G,18013,22,30915,5,18598,3,33879,38,44957,4,47.5,0,16,99,1,+++++,+++,255,0,253,4,+++++,+++,226,1 > > ---------8<---------8<---------patch---------8<---------8<--------- > > > > fs/nilfs2/ioctl.c | 12 ++++++------ > > 1 files changed, 6 insertions(+), 6 deletions(-) > > > > Signed-off-by: Jiro SEKIBA > > --- > > fs/nilfs2/ioctl.c | 12 ++++++------ > > 1 files changed, 6 insertions(+), 6 deletions(-) > > > > diff --git a/fs/nilfs2/ioctl.c b/fs/nilfs2/ioctl.c > > index 6ea5f87..83a885c 100644 > > --- a/fs/nilfs2/ioctl.c > > +++ b/fs/nilfs2/ioctl.c > > @@ -442,12 +442,6 @@ int nilfs_ioctl_prepare_clean_segments(struct the_nilfs *nilfs, > > const char *msg; > > int ret; > > > > - ret = nilfs_ioctl_move_blocks(nilfs, &argv[0], kbufs[0]); > > - if (ret < 0) { > > - msg = "cannot read source blocks"; > > - goto failed; > > - } > > - > > ret = nilfs_ioctl_delete_checkpoints(nilfs, &argv[1], kbufs[1]); > > if (ret < 0) { > > /* > > @@ -548,6 +542,12 @@ static int nilfs_ioctl_clean_segments(struct inode *inode, struct file *filp, > > } > > } > > > > + ret = nilfs_ioctl_move_blocks(nilfs, &argv[0], kbufs[0]); > > + if (ret < 0) { > > + printk(KERN_ERR "NILFS: GC failed during preparation: cannot read source blocks: err=%d\n", ret); > > + goto out_free; > > + } > > + > > ret = nilfs_clean_segments(inode->i_sb, argv, kbufs); > > > > out_free: > > -- > > 1.5.6.5 > _______________________________________________ > users mailing list > users-JrjvKiOkagjYtjvyW6yDsg@public.gmane.org > https://www.nilfs.org/mailman/listinfo/users > > > -- Jiro SEKIBA