ext4 data=writeback performs worse than data=ordered now

Quoted 6 lines

The worst case happens for the USB key, where both old/new kernels see ~10% worse performance for data=writeback.

ext4 ext4:wb
------------------------ ------------------------

                    6.20       -10.6%         5.54  fat/UKEY-thresh=100M/ext4-1dd-1-3.2.0-rc3-pause6+

Some more comparison numbers for the above worst case.

I don't see obvious differences from the balance_dirty_pages graphs,

Hi,

Shaohua recently found that ext4 writeback mode could perform worse than ordered mode in some cases. It may not be a big problem, however we'd like to share some information on our findings.

I tested both 3.2 and 3.1 kernels on normal SATA disks and USB key. The interesting thing is, data=writeback used to run a bit faster than data=ordered, however situation get inverted presumably by the IO-less dirty throttling.

Quoted 6 lines · Wu Fengguang

Hi,

Shaohua recently found that ext4 writeback mode could perform worse than ordered mode in some cases. It may not be a big problem, however we'd like to share some information on our findings.

Quoted 4 lines

I tested both 3.2 and 3.1 kernels on normal SATA disks and USB key. The interesting thing is, data=writeback used to run a bit faster than data=ordered, however situation get inverted presumably by the IO-less dirty throttling.

Interesting. What sort of workloads are you using to do these measurements? How many writer threads; I assume you are doing sequential writes which are extending one or more files, etc?

I suspect it's due to the throttling meaning that each thread is getting to send less data to the disk, and so there is more seeking going on with data=writeback, where as with data=ordered, at each journal commit we are forcing all of the dirty pages out to disk, one inode at a time, and this is resulting in a more efficient writeback compared to when the writeback code is getting to make its own choices about how much each inode gets to write out at at time.

It would be interesting to see what would happen if in ext4_da_writepages(), we completely ignore how many pages are requested to be written back by the writeback code, and just simply write back all of the dirty pages, and see if that brings the performance back.

Quoted 10 lines · Wu Fengguang

Hi,

Shaohua recently found that ext4 writeback mode could perform worse than ordered mode in some cases. It may not be a big problem, however we'd like to share some information on our findings.

I tested both 3.2 and 3.1 kernels on normal SATA disks and USB key. The interesting thing is, data=writeback used to run a bit faster than data=ordered, however situation get inverted presumably by the IO-less dirty throttling.

Interesting. What sort of workloads are you using to do these measurements? How many writer threads; I assume you are doing sequential writes which are extending one or more files, etc?

I suspect it's due to the throttling meaning that each thread is getting to send less data to the disk, and so there is more seeking going on with data=writeback, where as with data=ordered, at each journal commit we are forcing all of the dirty pages out to disk, one inode at a time, and this is resulting in a more efficient writeback compared to when the writeback code is getting to make its own choices about how much each inode gets to write out at at time.

It would be interesting to see what would happen if in ext4_da_writepages(), we completely ignore how many pages are requested to be written back by the writeback code, and just simply write back all of the dirty pages, and see if that brings the performance back.

Hi Ted/Fengguang,

Quoted 24 lines · Ted Ts'o

Quoted 10 lines · Wu Fengguang

Hi,

Shaohua recently found that ext4 writeback mode could perform worse than ordered mode in some cases. It may not be a big problem, however we'd like to share some information on our findings.

I tested both 3.2 and 3.1 kernels on normal SATA disks and USB key. The interesting thing is, data=writeback used to run a bit faster than data=ordered, however situation get inverted presumably by the IO-less dirty throttling.

Interesting. What sort of workloads are you using to do these measurements? How many writer threads; I assume you are doing sequential writes which are extending one or more files, etc?

I suspect it's due to the throttling meaning that each thread is getting to send less data to the disk, and so there is more seeking going on with data=writeback, where as with data=ordered, at each journal commit we are forcing all of the dirty pages out to disk, one inode at a time, and this is resulting in a more efficient writeback compared to when the writeback code is getting to make its own choices about how much each inode gets to write out at at time.

Quoted 5 lines

It would be interesting to see what would happen if in ext4_da_writepages(), we completely ignore how many pages are requested to be written back by the writeback code, and just simply write back all of the dirty pages, and see if that brings the performance back.

I guess fengguang's test is a buffer write dd test. Here we have found some performance regression from 18 because of the delayed allocation. In case of delayed allocation, we will create the extent tree during writepages which would delay the write because ext4_da_write_begin would down_read the i_data_sem to map the block while writepages would down_write it so we have seen some severe delay in ext4_da_write_begin (around 3s). And instead of increasing the page numbers of every writepages, some tests shows that the decrease makes the performance increase. I will dive into it soon to see what's going on there.

So Fengguang, would you please keep the page number in ext4_da_writepages passed by writeback(instead of the bumping) and check the result?

Quoted 10 lines · Wu Fengguang

Hi,

Shaohua recently found that ext4 writeback mode could perform worse than ordered mode in some cases. It may not be a big problem, however we'd like to share some information on our findings.

I tested both 3.2 and 3.1 kernels on normal SATA disks and USB key. The interesting thing is, data=writeback used to run a bit faster than data=ordered, however situation get inverted presumably by the IO-less dirty throttling.

Interesting. What sort of workloads are you using to do these measurements? How many writer threads; I assume you are doing sequential writes which are extending one or more files, etc?

I suspect it's due to the throttling meaning that each thread is getting to send less data to the disk, and so there is more seeking going on with data=writeback, where as with data=ordered, at each journal commit we are forcing all of the dirty pages out to disk, one inode at a time, and this is resulting in a more efficient writeback compared to when the writeback code is getting to make its own choices about how much each inode gets to write out at at time.

It would be interesting to see what would happen if in ext4_da_writepages(), we completely ignore how many pages are requested to be written back by the writeback code, and just simply write back all of the dirty pages, and see if that brings the performance back.

I found sometimes one disk hasn't any request inflight, but we can't send request to the disk, because the scsi host's resource (the queue depth) is used out, looks we send too many requests from other disks and leave some disks starved. The resource imbalance in scsi isn't a new problem, even 3.1 has such issue, so I'd think writeback introduces new imbalance between the 12 disks. In fact, if I limit disk's queue depth to 10, in this way the 12 disks will not impact each other in scsi layer, the performance regression fully disappears for both writeback and order mode.

Quoted 9 lines

I found sometimes one disk hasn't any request inflight, but we can't send request to the disk, because the scsi host's resource (the queue depth) is used out, looks we send too many requests from other disks and leave some disks starved. The resource imbalance in scsi isn't a new problem, even 3.1 has such issue, so I'd think writeback introduces new imbalance between the 12 disks. In fact, if I limit disk's queue depth to 10, in this way the 12 disks will not impact each other in scsi layer, the performance regression fully disappears for both writeback and order mode.

I observe similar issue in MD. The default

        q->nr_requests = BLKDEV_MAX_RQ;

is too small for large arrays, and I end up doing

        echo 1280 > /sys/block/md0/queue/nr_requests

in my tests.

Quoted 16 lines · On Thu, 15 Dec 2011 09:00:10 +0800 Wu Fengguang <[email protected]> wrote:

Quoted 9 lines

I found sometimes one disk hasn't any request inflight, but we can't send request to the disk, because the scsi host's resource (the queue depth) is used out, looks we send too many requests from other disks and leave some disks starved. The resource imbalance in scsi isn't a new problem, even 3.1 has such issue, so I'd think writeback introduces new imbalance between the 12 disks. In fact, if I limit disk's queue depth to 10, in this way the 12 disks will not impact each other in scsi layer, the performance regression fully disappears for both writeback and order mode.

I observe similar issue in MD. The default

        q->nr_requests = BLKDEV_MAX_RQ;

is too small for large arrays, and I end up doing

Quoted 3 lines

        echo 1280 > /sys/block/md0/queue/nr_requests

in my tests.

And you find this makes a difference?

That is very surprising because md devices don't use requests (and really use the 'queue' at all) and definitely don't make use of nr_requests.

Quoted 16 lines · On Thu, 15 Dec 2011 09:00:10 +0800 Wu Fengguang <[email protected]> wrote:

Quoted 9 lines

I found sometimes one disk hasn't any request inflight, but we can't send request to the disk, because the scsi host's resource (the queue depth) is used out, looks we send too many requests from other disks and leave some disks starved. The resource imbalance in scsi isn't a new problem, even 3.1 has such issue, so I'd think writeback introduces new imbalance between the 12 disks. In fact, if I limit disk's queue depth to 10, in this way the 12 disks will not impact each other in scsi layer, the performance regression fully disappears for both writeback and order mode.

I observe similar issue in MD. The default

        q->nr_requests = BLKDEV_MAX_RQ;

is too small for large arrays, and I end up doing

Quoted 3 lines

        echo 1280 > /sys/block/md0/queue/nr_requests

in my tests.

And you find this makes a difference?

That is very surprising because md devices don't use requests (and really use the 'queue' at all) and definitely don't make use of nr_requests.

Quoted 24 lines · Ted Ts'o

Quoted 10 lines · Wu Fengguang

Hi,

Shaohua recently found that ext4 writeback mode could perform worse than ordered mode in some cases. It may not be a big problem, however we'd like to share some information on our findings.

I tested both 3.2 and 3.1 kernels on normal SATA disks and USB key. The interesting thing is, data=writeback used to run a bit faster than data=ordered, however situation get inverted presumably by the IO-less dirty throttling.

Interesting. What sort of workloads are you using to do these measurements? How many writer threads; I assume you are doing sequential writes which are extending one or more files, etc?

I suspect it's due to the throttling meaning that each thread is getting to send less data to the disk, and so there is more seeking going on with data=writeback, where as with data=ordered, at each journal commit we are forcing all of the dirty pages out to disk, one inode at a time, and this is resulting in a more efficient writeback compared to when the writeback code is getting to make its own choices about how much each inode gets to write out at at time.

Quoted 5 lines

It would be interesting to see what would happen if in ext4_da_writepages(), we completely ignore how many pages are requested to be written back by the writeback code, and just simply write back all of the dirty pages, and see if that brings the performance back.

I saw the issue in a machine with a LSI 1068e HBA card and 12 disks. there is about 20% performance regression with data=writeback comparing 3.1 and 3.2-rc. with data=order, there is small regression too. Reverting writeback changes recover the regression for both cases.

My investigation shows the block size writing to disk isn't changed with data=writeback. The block size is still very big, 256k IIRC, which is the max block size in the disks. And I just have one thread for each disk, so seek definitely isn't a problem in my workload.

I found sometimes one disk hasn't any request inflight, but we can't send request to the disk, because the scsi host's resource (the queue depth) is used out, looks we send too many requests from other disks and leave some disks starved. The resource imbalance in scsi isn't a new

problem, even 3.1 has such issue, so I'd think writeback introduces new imbalance between the 12 disks. In fact, if I limit disk's queue depth to 10, in this way the 12 disks will not impact each other in scsi layer, the performance regression fully disappears for both writeback and order mode.

Thanks,
Shaohua

--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

Quoted 40 lines · Shaohua Li

Quoted 24 lines · Ted Ts'o

Quoted 10 lines · Wu Fengguang

Hi,

Shaohua recently found that ext4 writeback mode could perform worse than ordered mode in some cases. It may not be a big problem, however we'd like to share some information on our findings.

I tested both 3.2 and 3.1 kernels on normal SATA disks and USB key. The interesting thing is, data=writeback used to run a bit faster than data=ordered, however situation get inverted presumably by the IO-less dirty throttling.

Interesting. What sort of workloads are you using to do these measurements? How many writer threads; I assume you are doing sequential writes which are extending one or more files, etc?

I suspect it's due to the throttling meaning that each thread is getting to send less data to the disk, and so there is more seeking going on with data=writeback, where as with data=ordered, at each journal commit we are forcing all of the dirty pages out to disk, one inode at a time, and this is resulting in a more efficient writeback compared to when the writeback code is getting to make its own choices about how much each inode gets to write out at at time.

Quoted 5 lines

It would be interesting to see what would happen if in ext4_da_writepages(), we completely ignore how many pages are requested to be written back by the writeback code, and just simply write back all of the dirty pages, and see if that brings the performance back.

I saw the issue in a machine with a LSI 1068e HBA card and 12 disks. there is about 20% performance regression with data=writeback comparing 3.1 and 3.2-rc. with data=order, there is small regression too. Reverting writeback changes recover the regression for both cases.

My investigation shows the block size writing to disk isn't changed with data=writeback. The block size is still very big, 256k IIRC, which is the max block size in the disks. And I just have one thread for each disk, so seek definitely isn't a problem in my workload.

Quoted 4 lines

I found sometimes one disk hasn't any request inflight, but we can't send request to the disk, because the scsi host's resource (the queue depth) is used out, looks we send too many requests from other disks and leave some disks starved. The resource imbalance in scsi isn't a new

I wonder, does the patch in:
http://lkml.indiana.edu/hypermail/linux/kernel/1105.3/02339.html
help with this starvation problem? I noticed a similar problem and sent a patch, but LSI folks never responded. Maybe two complaining users can change that. The biggest MaxQ I've seen on LSI SAS is 511, and the driver clamps the value it passes to the SCSI layer to whatever the controller reports as its MaxQ (in /proc/mpt/summary).

Quoted 46 lines · Darrick J. Wong

Quoted 40 lines · Shaohua Li

Quoted 24 lines · Ted Ts'o

Quoted 10 lines · Wu Fengguang

Hi,

Shaohua recently found that ext4 writeback mode could perform worse than ordered mode in some cases. It may not be a big problem, however we'd like to share some information on our findings.

I tested both 3.2 and 3.1 kernels on normal SATA disks and USB key. The interesting thing is, data=writeback used to run a bit faster than data=ordered, however situation get inverted presumably by the IO-less dirty throttling.

Interesting. What sort of workloads are you using to do these measurements? How many writer threads; I assume you are doing sequential writes which are extending one or more files, etc?

I suspect it's due to the throttling meaning that each thread is getting to send less data to the disk, and so there is more seeking going on with data=writeback, where as with data=ordered, at each journal commit we are forcing all of the dirty pages out to disk, one inode at a time, and this is resulting in a more efficient writeback compared to when the writeback code is getting to make its own choices about how much each inode gets to write out at at time.

Quoted 5 lines

It would be interesting to see what would happen if in ext4_da_writepages(), we completely ignore how many pages are requested to be written back by the writeback code, and just simply write back all of the dirty pages, and see if that brings the performance back.

I saw the issue in a machine with a LSI 1068e HBA card and 12 disks. there is about 20% performance regression with data=writeback comparing 3.1 and 3.2-rc. with data=order, there is small regression too. Reverting writeback changes recover the regression for both cases.

My investigation shows the block size writing to disk isn't changed with data=writeback. The block size is still very big, 256k IIRC, which is the max block size in the disks. And I just have one thread for each disk, so seek definitely isn't a problem in my workload.

Quoted 4 lines

I found sometimes one disk hasn't any request inflight, but we can't send request to the disk, because the scsi host's resource (the queue depth) is used out, looks we send too many requests from other disks and leave some disks starved. The resource imbalance in scsi isn't a new

Quoted 7 lines

I wonder, does the patch in:
http://lkml.indiana.edu/hypermail/linux/kernel/1105.3/02339.html
help with this starvation problem? I noticed a similar problem and sent a patch, but LSI folks never responded. Maybe two complaining users can change that. The biggest MaxQ I've seen on LSI SAS is 511, and the driver clamps the value it passes to the SCSI layer to whatever the controller reports as its MaxQ (in /proc/mpt/summary).

this should recover the regression too. But I'm afraid it's just a workaround and will hide some issues. what if I have 120 disks instead of 12 disks? I observed one disk can burst 20 requests while the total the scsi host queue depth is 127, leaving other disks starved. I'm hoping to understand why there is such imbalance.

Thanks,
Shaohua

Quoted 54 lines · Shaohua Li

Quoted 46 lines · Darrick J. Wong

Quoted 40 lines · Shaohua Li

Quoted 24 lines · Ted Ts'o

Quoted 10 lines · Wu Fengguang

Hi,

Shaohua recently found that ext4 writeback mode could perform worse than ordered mode in some cases. It may not be a big problem, however we'd like to share some information on our findings.

I tested both 3.2 and 3.1 kernels on normal SATA disks and USB key. The interesting thing is, data=writeback used to run a bit faster than data=ordered, however situation get inverted presumably by the IO-less dirty throttling.

Interesting. What sort of workloads are you using to do these measurements? How many writer threads; I assume you are doing sequential writes which are extending one or more files, etc?

I suspect it's due to the throttling meaning that each thread is getting to send less data to the disk, and so there is more seeking going on with data=writeback, where as with data=ordered, at each journal commit we are forcing all of the dirty pages out to disk, one inode at a time, and this is resulting in a more efficient writeback compared to when the writeback code is getting to make its own choices about how much each inode gets to write out at at time.

Quoted 5 lines

It would be interesting to see what would happen if in ext4_da_writepages(), we completely ignore how many pages are requested to be written back by the writeback code, and just simply write back all of the dirty pages, and see if that brings the performance back.

I saw the issue in a machine with a LSI 1068e HBA card and 12 disks. there is about 20% performance regression with data=writeback comparing 3.1 and 3.2-rc. with data=order, there is small regression too. Reverting writeback changes recover the regression for both cases.

My investigation shows the block size writing to disk isn't changed with data=writeback. The block size is still very big, 256k IIRC, which is the max block size in the disks. And I just have one thread for each disk, so seek definitely isn't a problem in my workload.

Quoted 4 lines

I found sometimes one disk hasn't any request inflight, but we can't send request to the disk, because the scsi host's resource (the queue depth) is used out, looks we send too many requests from other disks and leave some disks starved. The resource imbalance in scsi isn't a new

Quoted 7 lines

I wonder, does the patch in:
http://lkml.indiana.edu/hypermail/linux/kernel/1105.3/02339.html
help with this starvation problem? I noticed a similar problem and sent a patch, but LSI folks never responded. Maybe two complaining users can change that. The biggest MaxQ I've seen on LSI SAS is 511, and the driver clamps the value it passes to the SCSI layer to whatever the controller reports as its MaxQ (in /proc/mpt/summary).

Quoted 5 lines

this should recover the regression too. But I'm afraid it's just a workaround and will hide some issues. what if I have 120 disks instead of 12 disks? I observed one disk can burst 20 requests while the total the scsi host queue depth is 127, leaving other disks starved. I'm hoping to understand why there is such imbalance.

<shrug> I didn't say it would /fix/ the imbalanced-starvation problem, but we might as well take full advantage of the hardware. Even if all it does is enable the user to plug in more disks before things get whacky, I was hoping that someone else could at least give it a spin and say "Yes, this does what it's alleged to do, and without breaking things". :)

afaict SCSI doesn't try to balance requests heading towards the HBA; it's all FCFS.