| Message ID | 20170911033845.53701-1-jaegeuk@kernel.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 2017/9/11 11:38, Jaegeuk Kim wrote: > If issue_cond is true, it does double count for # of issued commands. > > Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org> Reviewed-by: Chao Yu <yuchao0@huawei.com> > --- > fs/f2fs/segment.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c > index 7fd742f747ce..25196ff5d587 100644 > --- a/fs/f2fs/segment.c > +++ b/fs/f2fs/segment.c > @@ -1087,7 +1087,7 @@ static int __issue_discard_cmd(struct f2fs_sb_info *sbi, bool issue_cond) > issued++; > __submit_discard_cmd(sbi, dc); > } > - if (issue_cond && iter++ > DISCARD_ISSUE_RATE) > + if (issue_cond && iter > DISCARD_ISSUE_RATE) > goto out; > } > if (list_empty(pend_list) && dcc->pend_list_tag[i] & P_TRIM) >
On 2017/9/12 9:53, Chao Yu wrote: > On 2017/9/11 11:38, Jaegeuk Kim wrote: >> If issue_cond is true, it does double count for # of issued commands. >> >> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org> > > Reviewed-by: Chao Yu <yuchao0@huawei.com> As Daeho Jeong mentioned, the change makes 'iter > DISCARD_ISSUE_RATE' dead code, I just misread iter and issued variable, sorry. :( Thanks, > >> --- >> fs/f2fs/segment.c | 2 +- >> 1 file changed, 1 insertion(+), 1 deletion(-) >> >> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c >> index 7fd742f747ce..25196ff5d587 100644 >> --- a/fs/f2fs/segment.c >> +++ b/fs/f2fs/segment.c >> @@ -1087,7 +1087,7 @@ static int __issue_discard_cmd(struct f2fs_sb_info *sbi, bool issue_cond) >> issued++; >> __submit_discard_cmd(sbi, dc); >> } >> - if (issue_cond && iter++ > DISCARD_ISSUE_RATE) >> + if (issue_cond && iter > DISCARD_ISSUE_RATE) >> goto out; >> } >> if (list_empty(pend_list) && dcc->pend_list_tag[i] & P_TRIM) >> > > >
On 09/12, Chao Yu wrote: > On 2017/9/12 9:53, Chao Yu wrote: > > On 2017/9/11 11:38, Jaegeuk Kim wrote: > >> If issue_cond is true, it does double count for # of issued commands. > >> > >> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org> > > > > Reviewed-by: Chao Yu <yuchao0@huawei.com> > > As Daeho Jeong mentioned, the change makes 'iter > DISCARD_ISSUE_RATE' dead > code, I just misread iter and issued variable, sorry. :( Yeah, that's exactly like what I made a mistake before. I should have mentioned that earlier. :) > > Thanks, > > > > >> --- > >> fs/f2fs/segment.c | 2 +- > >> 1 file changed, 1 insertion(+), 1 deletion(-) > >> > >> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c > >> index 7fd742f747ce..25196ff5d587 100644 > >> --- a/fs/f2fs/segment.c > >> +++ b/fs/f2fs/segment.c > >> @@ -1087,7 +1087,7 @@ static int __issue_discard_cmd(struct f2fs_sb_info *sbi, bool issue_cond) > >> issued++; > >> __submit_discard_cmd(sbi, dc); > >> } > >> - if (issue_cond && iter++ > DISCARD_ISSUE_RATE) > >> + if (issue_cond && iter > DISCARD_ISSUE_RATE) > >> goto out; > >> } > >> if (list_empty(pend_list) && dcc->pend_list_tag[i] & P_TRIM) > >> > > > > > >
> Yeah, that's exactly like what I made a mistake before. > I should have mentioned that earlier. :) Or I think the previous code which used "iter++" might be right. You might just want to check the fixed number of small discards, DISCARD_ISSUE_RATE, when issue_cond is "true". Anyways, I have another question about this function. How about just issuing, not checking whether it is idle, the fixed number of small discards, DISCARD_ISSUE_RATE, when issue_cond is "true". Actually, the discard commands will be issued as "asynchronous" requests, which has a low priority in the I/O scheduler, so the performance degradation of other threads by doing this will not be much severe, but we can make the performance of the storage device better even if there is no idle. I am just worried about the storage device I/O performance gets worse under I/O intensive senario where there is no idle Thanks,
On 09/12, Daeho Jeong wrote: > > Yeah, that's exactly like what I made a mistake before. > > I should have mentioned that earlier. :) > > Or I think the previous code which used "iter++" might be right. > You might just want to check the fixed number of small discards, DISCARD_ISSUE_RATE, > when issue_cond is "true". > > Anyways, I have another question about this function. > How about just issuing, not checking whether it is idle, the fixed number of small > discards, DISCARD_ISSUE_RATE, when issue_cond is "true". > Actually, the discard commands will be issued as "asynchronous" requests, > which has a low priority in the I/O scheduler, > so the performance degradation of other threads by doing this will not be much severe, > but we can make the performance of the storage device better even if there is no idle. I don't think I/O scheduler can efficiently prioritize discard commands and user requests. The proper way that we can do would be waiting for idle time at this moment. Thanks, > > I am just worried about the storage device I/O performance gets worse > under I/O intensive senario where there is no idle > > Thanks,
On 2017/9/12 12:34, Daeho Jeong wrote: >> Yeah, that's exactly like what I made a mistake before. >> I should have mentioned that earlier. :) > > Or I think the previous code which used "iter++" might be right. > You might just want to check the fixed number of small discards, DISCARD_ISSUE_RATE, > when issue_cond is "true". > > Anyways, I have another question about this function. > How about just issuing, not checking whether it is idle, the fixed number of small > discards, DISCARD_ISSUE_RATE, when issue_cond is "true". > Actually, the discard commands will be issued as "asynchronous" requests, > which has a low priority in the I/O scheduler, > so the performance degradation of other threads by doing this will not be much severe, > but we can make the performance of the storage device better even if there is no idle. Actually, we didn't change priority of discard command, so that it is still synchronous IO for I/O scheduler, hence I/O interference will still exist if we try to issue discard without IO aware ability. Of course we can change the priority of discard command to lower, but potential issue is that with ROW I/O scheduler in kernel or FTL, async I/O will handle very slowly in heavy load scenario, if we are going to trigger sync write IO in place in where we're doing async discard, we will face long latency. Still I think it is worth to build the ability to issue async discard as a part of discard policy and later we can adjust policy based on different scenario. Thanks, > > I am just worried about the storage device I/O performance gets worse > under I/O intensive senario where there is no idle > > Thanks, > > ------------------------------------------------------------------------------ > Check out the vibrant tech community on one of the world's most > engaging tech sites, Slashdot.org! http://sdm.link/slashdot > _______________________________________________ > Linux-f2fs-devel mailing list > Linux-f2fs-devel@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel >
> Actually, we didn't change priority of discard command, so that it is still > synchronous IO for I/O scheduler, hence I/O interference will still exist if we > try to issue discard without IO aware ability. > Of course we can change the priority of discard command to lower, but potential > issue is that with ROW I/O scheduler in kernel or FTL, async I/O will handle > very slowly in heavy load scenario, if we are going to trigger sync write IO in > place in where we're doing async discard, we will face long latency. > Still I think it is worth to build the ability to issue async discard as a part > of discard policy and later we can adjust policy based on different scenario. > Thanks, Oh, I see. f2fs is sending discard requests as "sync" requests, I didn't know that. Right, I just though in case of CFQ I/O scheduler, but f2fs has to consider the other schedulers, but CFQ. Thanks, :)
diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c index 7fd742f747ce..25196ff5d587 100644 --- a/fs/f2fs/segment.c +++ b/fs/f2fs/segment.c @@ -1087,7 +1087,7 @@ static int __issue_discard_cmd(struct f2fs_sb_info *sbi, bool issue_cond) issued++; __submit_discard_cmd(sbi, dc); } - if (issue_cond && iter++ > DISCARD_ISSUE_RATE) + if (issue_cond && iter > DISCARD_ISSUE_RATE) goto out; } if (list_empty(pend_list) && dcc->pend_list_tag[i] & P_TRIM)
If issue_cond is true, it does double count for # of issued commands. Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org> --- fs/f2fs/segment.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)