Got the answer to my own question; posting here if someone else encounters the same problem. The issue is that the default stripe size in a cephfs mount is 4 MB. If you are doing small reads (like 4k reads in the test I posted) inside the file, you'll end up pulling at least 4MB to the client (and then discarding most of the pulled data) even if you set readahead to zero. So, the solution for us was to set a lower stripe size, which aligns better with our workloads. Thanks and Regards, Ashu Pachauri On Fri, Mar 10, 2023 at 9:41 PM Ashu Pachauri <ashu210890@gmail.com> wrote:

Also, I am able to reproduce the network read amplification when I try to do very small reads from larger files. e.g.

for i in $(seq 1 10000); do dd if=test_${i} of=/dev/null bs=5k count=10 done

This piece of code generates a network traffic of 3.3 GB while it actually reads approx 500 MB of data.

Thanks and Regards, Ashu Pachauri

On Fri, Mar 10, 2023 at 9:22 PM Ashu Pachauri <ashu210890@gmail.com> wrote:

...

We have an internal use case where we back the storage of a proprietary database by a shared file system. We noticed something very odd when testing some workload with a local block device backed file system vs cephfs. We noticed that the amount of network IO done by cephfs is almost double compared to the IO done in case of a local file system backed by an attached block device.

We also noticed that CephFS thrashes through the page cache very quickly compared to the amount of data being read and think that the two issues might be related. So, I wrote a simple test.

1. I wrote 10k files 400KB each using dd (approx 4 GB data). 2. I dropped the page cache completely. 3. I then read these files serially, again using dd. The page cache usage shot up to 39 GB for reading such a small amount of data.

Following is the code used to repro this in bash:

for i in $(seq 1 10000); do dd if=/dev/zero of=test_${i} bs=4k count=100 done

sync; echo 1 > /proc/sys/vm/drop_caches

for i in $(seq 1 10000); do dd if=test_${i} of=/dev/null bs=4k count=100 done

The ceph version being used is: ceph version 15.2.13 (c44bc49e7a57a87d84dfff2a077a2058aa2172e2) octopus (stable)

The ceph configs being overriden: WHO MASK LEVEL OPTION VALUE RO mon advanced auth_allow_insecure_global_id_reclaim false

mgr advanced mgr/balancer/mode upmap

mgr advanced mgr/dashboard/server_addr 127.0.0.1 * mgr advanced mgr/dashboard/server_port 8443 * mgr advanced mgr/dashboard/ssl false * mgr advanced mgr/prometheus/server_addr 0.0.0.0 * mgr advanced mgr/prometheus/server_port 9283 * osd advanced bluestore_compression_algorithm lz4

osd advanced bluestore_compression_mode aggressive osd advanced bluestore_throttle_bytes 536870912 osd advanced osd_max_backfills 3

osd advanced osd_op_num_threads_per_shard_ssd 8 * osd advanced osd_scrub_auto_repair true

mds advanced client_oc false

mds advanced client_readahead_max_bytes 4096

mds advanced client_readahead_max_periods 1

mds advanced client_readahead_min 0

mds basic mds_cache_memory_limit 21474836480 client advanced client_oc false

client advanced client_readahead_max_bytes 4096

client advanced client_readahead_max_periods 1

client advanced client_readahead_min 0

client advanced fuse_disable_pagecache false

The cephfs mount options (note that readahead was disabled for this test): /mnt/cephfs type ceph (rw,relatime,name=cephfs,secret=<hidden>,acl,rasize=0)

Any help or pointers are appreciated; this is a major performance issue for us.

Thanks and Regards, Ashu Pachauri

On 15/03/2023 17:20, Frank Schilder wrote:

Hi Ashu,

are you talking about the kernel client? I can't find "stripe size" anywhere in its mount-documentation. Could you possibly post exactly what you did? Mount fstab line, config setting?

There is no mount option to do this in both userspace and kernel clients. You need to change the file layout, which is (4MB stripe_unit, 1 stripe_count and 4MB object_size) by default, instead. Certainly with a smaller size of the stripe_unit will work. But IMO it will depend and be careful, changing the layout may cause other performance issues in some case, for example too small stripe_unit size may split the sync read into more osd requests to different OSDs. I will generate one patch to make the kernel client wiser instead of blindly setting it to stripe_unit always. Thanks - Xiubo

Thanks! ================= Frank Schilder AIT Risø Campus Bygning 109, rum S14

________________________________________ From: Ashu Pachauri <ashu210890@gmail.com> Sent: 14 March 2023 19:23:42 To: ceph-users@ceph.io Subject: [ceph-users] Re: CephFS thrashing through the page cache

Got the answer to my own question; posting here if someone else encounters the same problem. The issue is that the default stripe size in a cephfs mount is 4 MB. If you are doing small reads (like 4k reads in the test I posted) inside the file, you'll end up pulling at least 4MB to the client (and then discarding most of the pulled data) even if you set readahead to zero. So, the solution for us was to set a lower stripe size, which aligns better with our workloads.

Thanks and Regards, Ashu Pachauri

On Fri, Mar 10, 2023 at 9:41 PM Ashu Pachauri <ashu210890@gmail.com> wrote:

...

Also, I am able to reproduce the network read amplification when I try to do very small reads from larger files. e.g.

for i in $(seq 1 10000); do dd if=test_${i} of=/dev/null bs=5k count=10 done

This piece of code generates a network traffic of 3.3 GB while it actually reads approx 500 MB of data.

Thanks and Regards, Ashu Pachauri

On Fri, Mar 10, 2023 at 9:22 PM Ashu Pachauri <ashu210890@gmail.com> wrote:

...

We have an internal use case where we back the storage of a proprietary database by a shared file system. We noticed something very odd when testing some workload with a local block device backed file system vs cephfs. We noticed that the amount of network IO done by cephfs is almost double compared to the IO done in case of a local file system backed by an attached block device.

We also noticed that CephFS thrashes through the page cache very quickly compared to the amount of data being read and think that the two issues might be related. So, I wrote a simple test.

1. I wrote 10k files 400KB each using dd (approx 4 GB data). 2. I dropped the page cache completely. 3. I then read these files serially, again using dd. The page cache usage shot up to 39 GB for reading such a small amount of data.

Following is the code used to repro this in bash:

for i in $(seq 1 10000); do dd if=/dev/zero of=test_${i} bs=4k count=100 done

sync; echo 1 > /proc/sys/vm/drop_caches

for i in $(seq 1 10000); do dd if=test_${i} of=/dev/null bs=4k count=100 done

The ceph version being used is: ceph version 15.2.13 (c44bc49e7a57a87d84dfff2a077a2058aa2172e2) octopus (stable)

The ceph configs being overriden: WHO MASK LEVEL OPTION VALUE RO mon advanced auth_allow_insecure_global_id_reclaim false

mgr advanced mgr/balancer/mode upmap

mgr advanced mgr/dashboard/server_addr 127.0.0.1 * mgr advanced mgr/dashboard/server_port 8443 * mgr advanced mgr/dashboard/ssl false * mgr advanced mgr/prometheus/server_addr 0.0.0.0 * mgr advanced mgr/prometheus/server_port 9283 * osd advanced bluestore_compression_algorithm lz4

osd advanced bluestore_compression_mode aggressive osd advanced bluestore_throttle_bytes 536870912 osd advanced osd_max_backfills 3

osd advanced osd_op_num_threads_per_shard_ssd 8 * osd advanced osd_scrub_auto_repair true

mds advanced client_oc false

mds advanced client_readahead_max_bytes 4096

mds advanced client_readahead_max_periods 1

mds advanced client_readahead_min 0

mds basic mds_cache_memory_limit 21474836480 client advanced client_oc false

client advanced client_readahead_max_bytes 4096

client advanced client_readahead_max_periods 1

client advanced client_readahead_min 0

client advanced fuse_disable_pagecache false

The cephfs mount options (note that readahead was disabled for this test): /mnt/cephfs type ceph (rw,relatime,name=cephfs,secret=<hidden>,acl,rasize=0)

Any help or pointers are appreciated; this is a major performance issue for us.

Thanks and Regards, Ashu Pachauri

---

ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io

-- Best Regards, Xiubo Li (李秀波) Email: xiubli@redhat.com/xiubli@ibm.com Slack: @Xiubo Li

On 17/03/2023 16:55, Ashu Pachauri wrote:

Hi Xiubo,

As you have correctly pointed out, I was talking about the stipe_unit setting in the file layout configuration. Here is the documentation for that for anyone else's reference: https://docs.ceph.com/en/quincy/cephfs/file-layouts/

As with any RAID0 setup, the stripe_unit is definitely workload dependent. Our use case requires us to read somewhere from a few kilobytes to a few hundred kilobytes at once. Having a 4MB default stripe_unit definitely hurts quite a bit. We were able to achieve almost 2x improvement in terms of average latency and overall throughput (for useful data) by reducing the stripe_unit. The rule of thumb is that you want to align the stripe_unit to your most common IO size.

Yeah, IMO this is ugly anyway. This is not as wise as the mm layer.

...

BTW, have you tried to set 'rasize' option to a small size instead of 0 ? Won't this work ?

No this won't work. I have tried it already. Since rasize simply impacts readahead, your minimum io size to the cephfs client will still be at the maximum of (rasize, stripe_unit).  rasize is a useful configuration only if it is required to be larger than the stripe_unit, otherwise it's not. Also, it's worth pointing out that simply setting rasize is not sufficient; one needs to change the corresponding configurations that control maximum/minimum readahead for ceph clients.

Yeah, this should work for old kernels before the ceph_netfs_expand_readahead() being introduced. I will improve it next week. Thanks for your reporting about this. Thanks - Xiubo

Thanks and Regards, Ashu Pachauri

On Fri, Mar 17, 2023 at 2:14 PM Xiubo Li <xiubli@redhat.com> wrote:

On 15/03/2023 17:20, Frank Schilder wrote: > Hi Ashu, > > are you talking about the kernel client? I can't find "stripe size" anywhere in its mount-documentation. Could you possibly post exactly what you did? Mount fstab line, config setting?

There is no mount option to do this in both userspace and kernel clients. You need to change the file layout, which is (4MB stripe_unit, 1 stripe_count and 4MB object_size) by default, instead.

Certainly with a smaller size of the stripe_unit will work. But IMO it will depend and be careful, changing the layout may cause other performance issues in some case, for example too small stripe_unit size may split the sync read into more osd requests to different OSDs.

I will generate one patch to make the kernel client wiser instead of blindly setting it to stripe_unit always.

Thanks

- Xiubo

> > Thanks! > ================= > Frank Schilder > AIT Risø Campus > Bygning 109, rum S14 > > ________________________________________ > From: Ashu Pachauri <ashu210890@gmail.com> > Sent: 14 March 2023 19:23:42 > To: ceph-users@ceph.io > Subject: [ceph-users] Re: CephFS thrashing through the page cache > > Got the answer to my own question; posting here if someone else > encounters the same problem. The issue is that the default stripe size in a > cephfs mount is 4 MB. If you are doing small reads (like 4k reads in the > test I posted) inside the file, you'll end up pulling at least 4MB to the > client (and then discarding most of the pulled data) even if you set > readahead to zero. So, the solution for us was to set a lower stripe size, > which aligns better with our workloads. > > Thanks and Regards, > Ashu Pachauri > > > On Fri, Mar 10, 2023 at 9:41 PM Ashu Pachauri <ashu210890@gmail.com> wrote: > >> Also, I am able to reproduce the network read amplification when I try to >> do very small reads from larger files. e.g. >> >> for i in $(seq 1 10000); do >>    dd if=test_${i} of=/dev/null bs=5k count=10 >> done >> >> >> This piece of code generates a network traffic of 3.3 GB while it actually >> reads approx 500 MB of data. >> >> >> Thanks and Regards, >> Ashu Pachauri >> >> On Fri, Mar 10, 2023 at 9:22 PM Ashu Pachauri <ashu210890@gmail.com> >> wrote: >> >>> We have an internal use case where we back the storage of a proprietary >>> database by a shared file system. We noticed something very odd when >>> testing some workload with a local block device backed file system vs >>> cephfs. We noticed that the amount of network IO done by cephfs is almost >>> double compared to the IO done in case of a local file system backed by an >>> attached block device. >>> >>> We also noticed that CephFS thrashes through the page cache very quickly >>> compared to the amount of data being read and think that the two issues >>> might be related. So, I wrote a simple test. >>> >>> 1. I wrote 10k files 400KB each using dd (approx 4 GB data). >>> 2. I dropped the page cache completely. >>> 3. I then read these files serially, again using dd. The page cache usage >>> shot up to 39 GB for reading such a small amount of data. >>> >>> Following is the code used to repro this in bash: >>> >>> for i in $(seq 1 10000); do >>>    dd if=/dev/zero of=test_${i} bs=4k count=100 >>> done >>> >>> sync; echo 1 > /proc/sys/vm/drop_caches >>> >>> for i in $(seq 1 10000); do >>>    dd if=test_${i} of=/dev/null bs=4k count=100 >>> done >>> >>> >>> The ceph version being used is: >>> ceph version 15.2.13 (c44bc49e7a57a87d84dfff2a077a2058aa2172e2) octopus >>> (stable) >>> >>> The ceph configs being overriden: >>> WHO       MASK  LEVEL     OPTION                VALUE >>>       RO >>>    mon           advanced auth_allow_insecure_global_id_reclaim  false >>> >>>    mgr           advanced  mgr/balancer/mode                 upmap >>> >>>    mgr           advanced mgr/dashboard/server_addr >>>   127.0.0.1    * >>>    mgr           advanced mgr/dashboard/server_port              8443 >>>      * >>>    mgr           advanced  mgr/dashboard/ssl                 false >>>       * >>>    mgr           advanced mgr/prometheus/server_addr             0.0.0.0 >>>       * >>>    mgr           advanced mgr/prometheus/server_port             9283 >>>      * >>>    osd           advanced bluestore_compression_algorithm        lz4 >>> >>>    osd           advanced bluestore_compression_mode >>> aggressive >>>    osd           advanced bluestore_throttle_bytes >>> 536870912 >>>    osd           advanced  osd_max_backfills                 3 >>> >>>    osd           advanced osd_op_num_threads_per_shard_ssd       8 >>>       * >>>    osd           advanced  osd_scrub_auto_repair                 true >>> >>>    mds           advanced  client_oc                 false >>> >>>    mds           advanced client_readahead_max_bytes             4096 >>> >>>    mds           advanced client_readahead_max_periods           1 >>> >>>    mds           advanced  client_readahead_min                  0 >>> >>>    mds           basic     mds_cache_memory_limit >>> 21474836480 >>>    client        advanced  client_oc                 false >>> >>>    client        advanced client_readahead_max_bytes             4096 >>> >>>    client        advanced client_readahead_max_periods           1 >>> >>>    client        advanced  client_readahead_min                  0 >>> >>>    client        advanced fuse_disable_pagecache                 false >>> >>> >>> The cephfs mount options (note that readahead was disabled for this test): >>> /mnt/cephfs type ceph >>> (rw,relatime,name=cephfs,secret=<hidden>,acl,rasize=0) >>> >>> Any help or pointers are appreciated; this is a major performance issue >>> for us. >>> >>> >>> Thanks and Regards, >>> Ashu Pachauri >>> > _______________________________________________ > ceph-users mailing list -- ceph-users@ceph.io > To unsubscribe send an email to ceph-users-leave@ceph.io > _______________________________________________ > ceph-users mailing list -- ceph-users@ceph.io > To unsubscribe send an email to ceph-users-leave@ceph.io > -- Best Regards,

Xiubo Li (李秀波)

Email: xiubli@redhat.com/xiubli@ibm.com <http://xiubli@redhat.com/xiubli@ibm.com> Slack: @Xiubo Li

-- Best Regards, Xiubo Li (李秀波) Email:xiubli@redhat.com/xiubli@ibm.com Slack: @Xiubo Li

Hi Xiubo, Did you get a chance to work on this? I am curious to test out the improvements. Thanks and Regards, Ashu Pachauri On Fri, Mar 17, 2023 at 3:33 PM Frank Schilder <frans@dtu.dk> wrote:

Hi Ashu,

thanks for the clarification. That's not an option that is easy to change. I hope that the modifications to the fs clients Xiubo has in mind will improve that. Thanks for flagging this performance issue. Would be great if this becomes part of a test suite.

Best regards, ================= Frank Schilder AIT Risø Campus Bygning 109, rum S14

________________________________________ From: Ashu Pachauri <ashu210890@gmail.com> Sent: 17 March 2023 09:55:25 To: Xiubo Li Cc: Frank Schilder; ceph-users@ceph.io Subject: Re: [ceph-users] Re: CephFS thrashing through the page cache

Hi Xiubo,

As you have correctly pointed out, I was talking about the stipe_unit setting in the file layout configuration. Here is the documentation for that for anyone else's reference: https://docs.ceph.com/en/quincy/cephfs/file-layouts/

As with any RAID0 setup, the stripe_unit is definitely workload dependent. Our use case requires us to read somewhere from a few kilobytes to a few hundred kilobytes at once. Having a 4MB default stripe_unit definitely hurts quite a bit. We were able to achieve almost 2x improvement in terms of average latency and overall throughput (for useful data) by reducing the stripe_unit. The rule of thumb is that you want to align the stripe_unit to your most common IO size.

...

BTW, have you tried to set 'rasize' option to a small size instead of 0 ? Won't this work ?

No this won't work. I have tried it already. Since rasize simply impacts readahead, your minimum io size to the cephfs client will still be at the maximum of (rasize, stripe_unit). rasize is a useful configuration only if it is required to be larger than the stripe_unit, otherwise it's not. Also, it's worth pointing out that simply setting rasize is not sufficient; one needs to change the corresponding configurations that control maximum/minimum readahead for ceph clients.

Thanks and Regards, Ashu Pachauri

On Fri, Mar 17, 2023 at 2:14 PM Xiubo Li <xiubli@redhat.com<mailto: xiubli@redhat.com>> wrote:

On 15/03/2023 17:20, Frank Schilder wrote:

...

Hi Ashu,

are you talking about the kernel client? I can't find "stripe size" anywhere in its mount-documentation. Could you possibly post exactly what you did? Mount fstab line, config setting?

There is no mount option to do this in both userspace and kernel clients. You need to change the file layout, which is (4MB stripe_unit, 1 stripe_count and 4MB object_size) by default, instead.

Certainly with a smaller size of the stripe_unit will work. But IMO it will depend and be careful, changing the layout may cause other performance issues in some case, for example too small stripe_unit size may split the sync read into more osd requests to different OSDs.

I will generate one patch to make the kernel client wiser instead of blindly setting it to stripe_unit always.

Thanks

- Xiubo

...

Thanks! ================= Frank Schilder AIT Risø Campus Bygning 109, rum S14

________________________________________ From: Ashu Pachauri <ashu210890@gmail.com<mailto:ashu210890@gmail.com>> Sent: 14 March 2023 19:23:42 To: ceph-users@ceph.io<mailto:ceph-users@ceph.io> Subject: [ceph-users] Re: CephFS thrashing through the page cache

Got the answer to my own question; posting here if someone else encounters the same problem. The issue is that the default stripe size

in a

...

cephfs mount is 4 MB. If you are doing small reads (like 4k reads in the test I posted) inside the file, you'll end up pulling at least 4MB to the client (and then discarding most of the pulled data) even if you set readahead to zero. So, the solution for us was to set a lower stripe size, which aligns better with our workloads.

Thanks and Regards, Ashu Pachauri

On Fri, Mar 10, 2023 at 9:41 PM Ashu Pachauri <ashu210890@gmail.com <mailto:ashu210890@gmail.com>> wrote:

...

Also, I am able to reproduce the network read amplification when I try to do very small reads from larger files. e.g.

for i in $(seq 1 10000); do dd if=test_${i} of=/dev/null bs=5k count=10 done

This piece of code generates a network traffic of 3.3 GB while it actually reads approx 500 MB of data.

Thanks and Regards, Ashu Pachauri

On Fri, Mar 10, 2023 at 9:22 PM Ashu Pachauri <ashu210890@gmail.com <mailto:ashu210890@gmail.com>> wrote:

...

We have an internal use case where we back the storage of a proprietary database by a shared file system. We noticed something very odd when testing some workload with a local block device backed file system vs cephfs. We noticed that the amount of network IO done by cephfs is almost double compared to the IO done in case of a local file system backed by an attached block device.

We also noticed that CephFS thrashes through the page cache very quickly compared to the amount of data being read and think that the two issues might be related. So, I wrote a simple test.

1. I wrote 10k files 400KB each using dd (approx 4 GB data). 2. I dropped the page cache completely. 3. I then read these files serially, again using dd. The page cache usage shot up to 39 GB for reading such a small amount of data.

Following is the code used to repro this in bash:

for i in $(seq 1 10000); do dd if=/dev/zero of=test_${i} bs=4k count=100 done

sync; echo 1 > /proc/sys/vm/drop_caches

for i in $(seq 1 10000); do dd if=test_${i} of=/dev/null bs=4k count=100 done

The ceph version being used is: ceph version 15.2.13 (c44bc49e7a57a87d84dfff2a077a2058aa2172e2) octopus (stable)

The ceph configs being overriden: WHO MASK LEVEL OPTION VALUE RO mon advanced auth_allow_insecure_global_id_reclaim false

mgr advanced mgr/balancer/mode upmap

mgr advanced mgr/dashboard/server_addr 127.0.0.1 * mgr advanced mgr/dashboard/server_port 8443 * mgr advanced mgr/dashboard/ssl false * mgr advanced mgr/prometheus/server_addr 0.0.0.0 * mgr advanced mgr/prometheus/server_port 9283 * osd advanced bluestore_compression_algorithm lz4

osd advanced bluestore_compression_mode aggressive osd advanced bluestore_throttle_bytes 536870912 osd advanced osd_max_backfills 3

osd advanced osd_op_num_threads_per_shard_ssd 8 * osd advanced osd_scrub_auto_repair true

mds advanced client_oc false

mds advanced client_readahead_max_bytes 4096

mds advanced client_readahead_max_periods 1

mds advanced client_readahead_min 0

mds basic mds_cache_memory_limit 21474836480 client advanced client_oc false

client advanced client_readahead_max_bytes 4096

client advanced client_readahead_max_periods 1

client advanced client_readahead_min 0

client advanced fuse_disable_pagecache false

The cephfs mount options (note that readahead was disabled for this test): /mnt/cephfs type ceph (rw,relatime,name=cephfs,secret=<hidden>,acl,rasize=0)

Any help or pointers are appreciated; this is a major performance issue for us.

Thanks and Regards, Ashu Pachauri

---

ceph-users mailing list -- ceph-users@ceph.io<mailto:ceph-users@ceph.io> To unsubscribe send an email to ceph-users-leave@ceph.io<mailto: ceph-users-leave@ceph.io> _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io<mailto:ceph-users@ceph.io> To unsubscribe send an email to ceph-users-leave@ceph.io<mailto: ceph-users-leave@ceph.io>

-- Best Regards,

Xiubo Li (李秀波)

Email: xiubli@redhat.com/xiubli@ibm.com< http://xiubli@redhat.com/xiubli@ibm.com> Slack: @Xiubo Li

On Fri, Mar 17, 2023 at 1:56 AM Ashu Pachauri <ashu210890@gmail.com> wrote:

Hi Xiubo,

As you have correctly pointed out, I was talking about the stipe_unit setting in the file layout configuration. Here is the documentation for that for anyone else's reference: https://docs.ceph.com/en/quincy/cephfs/file-layouts/

As with any RAID0 setup, the stripe_unit is definitely workload dependent. Our use case requires us to read somewhere from a few kilobytes to a few hundred kilobytes at once. Having a 4MB default stripe_unit definitely hurts quite a bit. We were able to achieve almost 2x improvement in terms of average latency and overall throughput (for useful data) by reducing the stripe_unit. The rule of thumb is that you want to align the stripe_unit to your most common IO size.

There's a lot more that goes into the stripe_unit than IO size for CephFS. This may improve your workload, but it more generally just means your IO accesses go out to more objects (so more PGs and more OSDs) generally. That can be good if you have low client counts doing random IO, but means that in general readahead (when not broken, as it appears to be in your kernel! :/) is much much much less effective and more expensive. -Greg

...

BTW, have you tried to set 'rasize' option to a small size instead of 0 ? Won't this work ?

No this won't work. I have tried it already. Since rasize simply impacts readahead, your minimum io size to the cephfs client will still be at the maximum of (rasize, stripe_unit). rasize is a useful configuration only if it is required to be larger than the stripe_unit, otherwise it's not. Also, it's worth pointing out that simply setting rasize is not sufficient; one needs to change the corresponding configurations that control maximum/minimum readahead for ceph clients.

Thanks and Regards, Ashu Pachauri

On Fri, Mar 17, 2023 at 2:14 PM Xiubo Li <xiubli@redhat.com> wrote:

...

On 15/03/2023 17:20, Frank Schilder wrote:

...

Hi Ashu,

are you talking about the kernel client? I can't find "stripe size" anywhere in its mount-documentation. Could you possibly post exactly what you did? Mount fstab line, config setting?

There is no mount option to do this in both userspace and kernel clients. You need to change the file layout, which is (4MB stripe_unit, 1 stripe_count and 4MB object_size) by default, instead.

Certainly with a smaller size of the stripe_unit will work. But IMO it will depend and be careful, changing the layout may cause other performance issues in some case, for example too small stripe_unit size may split the sync read into more osd requests to different OSDs.

I will generate one patch to make the kernel client wiser instead of blindly setting it to stripe_unit always.

Thanks

- Xiubo

...

Thanks! ================= Frank Schilder AIT Risø Campus Bygning 109, rum S14

________________________________________ From: Ashu Pachauri <ashu210890@gmail.com> Sent: 14 March 2023 19:23:42 To: ceph-users@ceph.io Subject: [ceph-users] Re: CephFS thrashing through the page cache

Got the answer to my own question; posting here if someone else encounters the same problem. The issue is that the default stripe size

in a

...

cephfs mount is 4 MB. If you are doing small reads (like 4k reads in the test I posted) inside the file, you'll end up pulling at least 4MB to the client (and then discarding most of the pulled data) even if you set readahead to zero. So, the solution for us was to set a lower stripe size, which aligns better with our workloads.

Thanks and Regards, Ashu Pachauri

On Fri, Mar 10, 2023 at 9:41 PM Ashu Pachauri <ashu210890@gmail.com> wrote:

...

Also, I am able to reproduce the network read amplification when I try to do very small reads from larger files. e.g.

for i in $(seq 1 10000); do dd if=test_${i} of=/dev/null bs=5k count=10 done

This piece of code generates a network traffic of 3.3 GB while it actually reads approx 500 MB of data.

Thanks and Regards, Ashu Pachauri

On Fri, Mar 10, 2023 at 9:22 PM Ashu Pachauri <ashu210890@gmail.com> wrote:

...

We have an internal use case where we back the storage of a proprietary database by a shared file system. We noticed something very odd when testing some workload with a local block device backed file system vs cephfs. We noticed that the amount of network IO done by cephfs is almost double compared to the IO done in case of a local file system backed by an attached block device.

We also noticed that CephFS thrashes through the page cache very quickly compared to the amount of data being read and think that the two issues might be related. So, I wrote a simple test.

1. I wrote 10k files 400KB each using dd (approx 4 GB data). 2. I dropped the page cache completely. 3. I then read these files serially, again using dd. The page cache usage shot up to 39 GB for reading such a small amount of data.

Following is the code used to repro this in bash:

for i in $(seq 1 10000); do dd if=/dev/zero of=test_${i} bs=4k count=100 done

sync; echo 1 > /proc/sys/vm/drop_caches

for i in $(seq 1 10000); do dd if=test_${i} of=/dev/null bs=4k count=100 done

The ceph version being used is: ceph version 15.2.13 (c44bc49e7a57a87d84dfff2a077a2058aa2172e2) octopus (stable)

The ceph configs being overriden: WHO MASK LEVEL OPTION VALUE RO mon advanced auth_allow_insecure_global_id_reclaim false

mgr advanced mgr/balancer/mode upmap

mgr advanced mgr/dashboard/server_addr 127.0.0.1 * mgr advanced mgr/dashboard/server_port 8443 * mgr advanced mgr/dashboard/ssl false * mgr advanced mgr/prometheus/server_addr 0.0.0.0 * mgr advanced mgr/prometheus/server_port 9283 * osd advanced bluestore_compression_algorithm lz4

osd advanced bluestore_compression_mode aggressive osd advanced bluestore_throttle_bytes 536870912 osd advanced osd_max_backfills 3

osd advanced osd_op_num_threads_per_shard_ssd 8 * osd advanced osd_scrub_auto_repair true

mds advanced client_oc false

mds advanced client_readahead_max_bytes 4096

mds advanced client_readahead_max_periods 1

mds advanced client_readahead_min 0

mds basic mds_cache_memory_limit 21474836480 client advanced client_oc false

client advanced client_readahead_max_bytes 4096

client advanced client_readahead_max_periods 1

client advanced client_readahead_min 0

client advanced fuse_disable_pagecache false

The cephfs mount options (note that readahead was disabled for this test): /mnt/cephfs type ceph (rw,relatime,name=cephfs,secret=<hidden>,acl,rasize=0)

Any help or pointers are appreciated; this is a major performance issue for us.

Thanks and Regards, Ashu Pachauri

---

ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io

-- Best Regards,

Xiubo Li (李秀波)

Email: xiubli@redhat.com/xiubli@ibm.com Slack: @Xiubo Li

_______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io

After going through the kernel code, the 'rasize' option seem never making much sense any more after netfs and 'ceph_netfs_expand_readahead()' were introduced to kernel ceph. The ceph client will try to expand the readahead windows twice and the second time it will always align the len down and up to the bounders of the 'stripe_unit'. Anyway this is unfriendly to small size reading case. Thanks - Xiubo On 15/03/2023 02:23, Ashu Pachauri wrote:

Got the answer to my own question; posting here if someone else encounters the same problem. The issue is that the default stripe size in a cephfs mount is 4 MB. If you are doing small reads (like 4k reads in the test I posted) inside the file, you'll end up pulling at least 4MB to the client (and then discarding most of the pulled data) even if you set readahead to zero. So, the solution for us was to set a lower stripe size, which aligns better with our workloads.

Thanks and Regards, Ashu Pachauri

On Fri, Mar 10, 2023 at 9:41 PM Ashu Pachauri <ashu210890@gmail.com> wrote:

...

Also, I am able to reproduce the network read amplification when I try to do very small reads from larger files. e.g.

for i in $(seq 1 10000); do dd if=test_${i} of=/dev/null bs=5k count=10 done

This piece of code generates a network traffic of 3.3 GB while it actually reads approx 500 MB of data.

Thanks and Regards, Ashu Pachauri

On Fri, Mar 10, 2023 at 9:22 PM Ashu Pachauri <ashu210890@gmail.com> wrote:

...

We have an internal use case where we back the storage of a proprietary database by a shared file system. We noticed something very odd when testing some workload with a local block device backed file system vs cephfs. We noticed that the amount of network IO done by cephfs is almost double compared to the IO done in case of a local file system backed by an attached block device.

We also noticed that CephFS thrashes through the page cache very quickly compared to the amount of data being read and think that the two issues might be related. So, I wrote a simple test.

1. I wrote 10k files 400KB each using dd (approx 4 GB data). 2. I dropped the page cache completely. 3. I then read these files serially, again using dd. The page cache usage shot up to 39 GB for reading such a small amount of data.

Following is the code used to repro this in bash:

for i in $(seq 1 10000); do dd if=/dev/zero of=test_${i} bs=4k count=100 done

sync; echo 1 > /proc/sys/vm/drop_caches

for i in $(seq 1 10000); do dd if=test_${i} of=/dev/null bs=4k count=100 done

The ceph version being used is: ceph version 15.2.13 (c44bc49e7a57a87d84dfff2a077a2058aa2172e2) octopus (stable)

The ceph configs being overriden: WHO MASK LEVEL OPTION VALUE RO mon advanced auth_allow_insecure_global_id_reclaim false

mgr advanced mgr/balancer/mode upmap

mgr advanced mgr/dashboard/server_addr 127.0.0.1 * mgr advanced mgr/dashboard/server_port 8443 * mgr advanced mgr/dashboard/ssl false * mgr advanced mgr/prometheus/server_addr 0.0.0.0 * mgr advanced mgr/prometheus/server_port 9283 * osd advanced bluestore_compression_algorithm lz4

osd advanced bluestore_compression_mode aggressive osd advanced bluestore_throttle_bytes 536870912 osd advanced osd_max_backfills 3

osd advanced osd_op_num_threads_per_shard_ssd 8 * osd advanced osd_scrub_auto_repair true

mds advanced client_oc false

mds advanced client_readahead_max_bytes 4096

mds advanced client_readahead_max_periods 1

mds advanced client_readahead_min 0

mds basic mds_cache_memory_limit 21474836480 client advanced client_oc false

client advanced client_readahead_max_bytes 4096

client advanced client_readahead_max_periods 1

client advanced client_readahead_min 0

client advanced fuse_disable_pagecache false

The cephfs mount options (note that readahead was disabled for this test): /mnt/cephfs type ceph (rw,relatime,name=cephfs,secret=<hidden>,acl,rasize=0)

Any help or pointers are appreciated; this is a major performance issue for us.

Thanks and Regards, Ashu Pachauri

---

ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io

-- Best Regards, Xiubo Li (李秀波) Email: xiubli@redhat.com/xiubli@ibm.com Slack: @Xiubo Li