Re: How to increase the size of requests written to a ceph image

[Jean-Charles Lopez <jelopez@xxxxxxxxxx>]

Hi Russell,

as you have 4 servers, assuming you are not doing EC pools, just stop all the OSDs on the second questionable server, mark the OSDs on that server as out, let the cluster rebalance and when all PGs are active+clean just replay the test.

All IOs should then go only to the other 3 servers.

JC

On Oct 19, 2017, at 13:49, Russell Glaue <rglaue@xxxxxxxx> wrote:

No, I have not ruled out the disk controller and backplane making the disks slower.

Is there a way I could test that theory, other than swapping out hardware?

-RG

On Thu, Oct 19, 2017 at 3:44 PM, David Turner <drakonstein@xxxxxxxxx> wrote:

Have you ruled out the disk controller and backplane in the server running slower?

On Thu, Oct 19, 2017 at 4:42 PM Russell Glaue <rglaue@xxxxxxxx> wrote:

I ran the test on the Ceph pool, and ran atop on all 4 storage servers, as suggested.

Out of the 4 servers:

3 of them performed with 17% to 30% disk %busy, and 11% CPU wait. Momentarily spiking up to 50% on one server, and 80% on another

The 2nd newest server was almost averaging 90% disk %busy and 150% CPU wait. And more than momentarily spiking to 101% disk busy and 250% CPU wait.

For this 2nd newest server, this was the statistics for about 8 of 9 disks, with the 9th disk not far behind the others.

I cannot believe all 9 disks are bad

They are the same disks as the newest 1st server, Crucial_CT960M500SSD1, and same exact server hardware too.

They were purchased at the same time in the same purchase order and arrived at the same time.

So I cannot believe I just happened to put 9 bad disks in one server, and 9 good ones in the other.

I know I have Ceph configured exactly the same on all servers

And I am sure I have the hardware settings configured exactly the same on the 1st and 2nd servers.

So if I were someone else, I would say it maybe is bad hardware on the 2nd server.

But the 2nd server is running very well without any hint of a problem.

Any other ideas or suggestions?

-RG

On Wed, Oct 18, 2017 at 3:40 PM, Maged Mokhtar <mmokhtar@xxxxxxxxxxx> wrote:

just run the same 32 threaded rados test as you did before and this time run atop while the test is running looking for %busy of cpu/disks. It should give an idea if there is a bottleneck in them. 

On 2017-10-18 21:35, Russell Glaue wrote:

I cannot run the write test reviewed at the ceph-how-to-test-if-your-ssd-is-suitable-as-a-journal-device blog. The tests write directly to the raw disk device.

Reading an infile (created with urandom) on one SSD, writing the outfile to another osd, yields about 17MB/s.

But Isn't this write speed limited by the speed in which in the dd infile can be read?

And I assume the best test should be run with no other load.

 

How does one run the rados bench "as stress"?

 

-RG

 

On Wed, Oct 18, 2017 at 1:33 PM, Maged Mokhtar <mmokhtar@xxxxxxxxxxx> wrote:

measuring resource load as outlined earlier will show if the drives are performing well or not. Also how many osds do you have  ?

On 2017-10-18 19:26, Russell Glaue wrote:

The SSD drives are Crucial M500

A Ceph user did some benchmarks and found it had good performance

https://forum.proxmox.com/threads/ceph-bad-performance-in-qemu-guests.21551/

 

However, a user comment from 3 years ago on the blog post you linked to says to avoid the Crucial M500

 

Yet, this performance posting tells that the Crucial M500 is good.

https://inside.servers.com/ssd-performance-2017-c4307a92dea

On Wed, Oct 18, 2017 at 11:53 AM, Maged Mokhtar <mmokhtar@xxxxxxxxxxx> wrote:

Check out the following link: some SSDs perform bad in Ceph due to sync writes to journal

https://www.sebastien-han.fr/blog/2014/10/10/ceph-how-to-test-if-your-ssd-is-suitable-as-a-journal-device/

Anther thing that can help is to re-run the rados 32 threads as stress and view resource usage using atop (or collectl/sar) to check for %busy cpu and %busy disks to give you an idea of what is holding down your cluster..for example: if cpu/disk % are all low then check your network/switches.  If disk %busy is high (90%) for all disks then your disks are the bottleneck: which either means you have SSDs that are not suitable for Ceph or you have too few disks (which i doubt is the case). If only 1 disk %busy is high, there may be something wrong with this disk should be removed.

Maged

On 2017-10-18 18:13, Russell Glaue wrote:

In my previous post, in one of my points I was wondering if the request size would increase if I enabled jumbo packets. currently it is disabled.

 

@jdillama: The qemu settings for both these two guest machines, with RAID/LVM and Ceph/rbd images, are the same. I am not thinking that changing the qemu settings of "min_io_size=<limited to 16bits>,opt_io_size=<RBD image object size>" will directly address the issue.

 

@mmokhtar: Ok. So you suggest the request size is the result of the problem and not the cause of the problem. meaning I should go after a different issue.

 

I have been trying to get write speeds up to what people on this mail list are discussing.

It seems that for our configuration, as it matches others, we should be getting about 70MB/s write speed.

But we are not getting that.

Single writes to disk are lucky to get 5MB/s to 6MB/s, but are typically 1MB/s to 2MB/s.

Monitoring the entire Ceph cluster (using http://cephdash.crapworks.de/), I have seen very rare momentary spikes up to 30MB/s.

 

My storage network is connected via a 10Gb switch

I have 4 storage servers with a LSI Logic MegaRAID SAS 2208 controller

Each storage server has 9 1TB SSD drives, each drive as 1 osd (no RAID)

Each drive is one LVM group, with two volumes - one volume for the osd, one volume for the journal

Each osd is formatted with xfs

The crush map is simple: default->rack->[host[1..4]->osd] with an evenly distributed weight

The redundancy is triple replication

 

While I have read comments that having the osd and journal on the same disk decreases write speed, I have also read that once past 8 OSDs per node this is the recommended configuration, however this is also the reason why SSD drives are used exclusively for OSDs in the storage nodes.

None-the-less, I was still expecting write speeds to be above 30MB/s, not below 6MB/s.

Even at 12x slower than the RAID, using my previously posted iostat data set, I should be seeing write speeds that average 10MB/s, not 2MB/s.

 

In regards to the rados benchmark tests you asked me to run, here is the output:

 

[centos7]# rados bench -p scbench -b 4096 30 write -t 1

Maintaining 1 concurrent writes of 4096 bytes to objects of size 4096 for up to 30 seconds or 0 objects

Object prefix: benchmark_data_hamms.sys.cu.cait.org_85049

  sec Cur ops   started  finished  avg MB/s  cur MB/s last lat(s)  avg lat(s)

    0       0         0         0         0         0           -           0

    1       1       201       200   0.78356   0.78125  0.00522307  0.00496574

    2       1       469       468  0.915303   1.04688  0.00437497  0.00426141

    3       1       741       740  0.964371    1.0625  0.00512853   0.0040434

    4       1       888       887  0.866739  0.574219  0.00307699  0.00450177

    5       1      1147      1146  0.895725   1.01172  0.00376454   0.0043559

    6       1      1325      1324  0.862293  0.695312  0.00459443    0.004525

    7       1      1494      1493   0.83339  0.660156  0.00461002  0.00458452

    8       1      1736      1735  0.847369  0.945312  0.00253971  0.00460458

    9       1      1998      1997  0.866922   1.02344  0.00236573  0.00450172

   10       1      2260      2259  0.882563   1.02344  0.00262179  0.00442152

   11       1      2526      2525  0.896775   1.03906  0.00336914  0.00435092

   12       1      2760      2759  0.898203  0.914062  0.00351827  0.00434491

   13       1      3016      3015  0.906025         1  0.00335703  0.00430691

   14       1      3257      3256  0.908545  0.941406  0.00332344  0.00429495

   15       1      3490      3489  0.908644  0.910156  0.00318815  0.00426387

   16       1      3728      3727  0.909952  0.929688   0.0032881  0.00428895

   17       1      3986      3985  0.915703   1.00781  0.00274809   0.0042614

   18       1      4250      4249  0.922116   1.03125  0.00287411  0.00423214

   19       1      4505      4504  0.926003  0.996094  0.00375435  0.00421442

2017-10-18 10:56:31.267173 min lat: 0.00181259 max lat: 0.270553 avg lat: 0.00420118

  sec Cur ops   started  finished  avg MB/s  cur MB/s last lat(s)  avg lat(s)

   20       1      4757      4756  0.928915  0.984375  0.00463972  0.00420118

   21       1      5009      5008   0.93155  0.984375  0.00360065  0.00418937

   22       1      5235      5234  0.929329  0.882812  0.00626214    0.004199

   23       1      5500      5499  0.933925   1.03516  0.00466584  0.00417836

   24       1      5708      5707  0.928861    0.8125  0.00285727  0.00420146

   25       0      5964      5964  0.931858   1.00391  0.00417383   0.0041881

   26       1      6216      6215  0.933722  0.980469   0.0041009  0.00417915

   27       1      6481      6480  0.937474   1.03516  0.00307484  0.00416118

   28       1      6745      6744  0.940819   1.03125  0.00266329  0.00414777

   29       1      7003      7002  0.943124   1.00781  0.00305905  0.00413758

   30       1      7271      7270  0.946578   1.04688  0.00391017  0.00412238

Total time run:         30.006060

Total writes made:      7272

Write size:             4096

Object size:            4096

Bandwidth (MB/sec):     0.946684

Stddev Bandwidth:       0.123762

Max bandwidth (MB/sec): 1.0625

Min bandwidth (MB/sec): 0.574219

Average IOPS:           242

Stddev IOPS:            31

Max IOPS:               272

Min IOPS:               147

Average Latency(s):     0.00412247

Stddev Latency(s):      0.00648437

Max latency(s):         0.270553

Min latency(s):         0.00175318

Cleaning up (deleting benchmark objects)

Clean up completed and total clean up time :29.069423

 

[centos7]# rados bench -p scbench -b 4096 30 write -t 32

Maintaining 32 concurrent writes of 4096 bytes to objects of size 4096 for up to 30 seconds or 0 objects

Object prefix: benchmark_data_hamms.sys.cu.cait.org_86076

  sec Cur ops   started  finished  avg MB/s  cur MB/s last lat(s)  avg lat(s)

    0       0         0         0         0         0           -           0

    1      32      3013      2981   11.6438   11.6445  0.00247906  0.00572026

    2      32      5349      5317   10.3834     9.125  0.00246662  0.00932016

    3      32      5707      5675    7.3883   1.39844  0.00389774   0.0156726

    4      32      5895      5863   5.72481  0.734375     1.13137   0.0167946

    5      32      6869      6837   5.34068   3.80469   0.0027652   0.0226577

    6      32      8901      8869   5.77306    7.9375   0.0053211   0.0216259

    7      32     10800     10768   6.00785   7.41797  0.00358187   0.0207418

    8      32     11825     11793   5.75728   4.00391  0.00217575   0.0215494

    9      32     12941     12909    5.6019   4.35938  0.00278512   0.0220567

   10      32     13317     13285   5.18849   1.46875   0.0034973   0.0240665

   11      32     16189     16157   5.73653   11.2188  0.00255841   0.0212708

   12      32     16749     16717   5.44077    2.1875  0.00330334   0.0215915

   13      32     16756     16724   5.02436 0.0273438  0.00338994    0.021849

   14      32     17908     17876   4.98686       4.5  0.00402598   0.0244568

   15      32     17936     17904   4.66171  0.109375  0.00375799   0.0245545

   16      32     18279     18247   4.45409   1.33984  0.00483873   0.0267929

   17      32     18372     18340   4.21346  0.363281  0.00505187   0.0275887

   18      32     19403     19371   4.20309   4.02734  0.00545154    0.029348

   19      31     19845     19814   4.07295   1.73047  0.00254726   0.0306775

2017-10-18 10:57:58.160536 min lat: 0.0015005 max lat: 2.27707 avg lat: 0.0307559

  sec Cur ops   started  finished  avg MB/s  cur MB/s last lat(s)  avg lat(s)

   20      31     20401     20370   3.97788   2.17188  0.00307238   0.0307559

   21      32     21338     21306   3.96254   3.65625  0.00464563   0.0312288

   22      32     23057     23025    4.0876   6.71484  0.00296295   0.0299267

   23      32     23057     23025   3.90988         0           -   0.0299267

   24      32     23803     23771   3.86837   1.45703  0.00301471   0.0312804

   25      32     24112     24080   3.76191   1.20703  0.00191063   0.0331462

   26      31     25303     25272   3.79629   4.65625  0.00794399   0.0329129

   27      32     28803     28771   4.16183    13.668   0.0109817   0.0297469

   28      32     29592     29560   4.12325   3.08203  0.00188185   0.0301911

   29      32     30595     30563   4.11616   3.91797  0.00379099   0.0296794

   30      32     31031     30999   4.03572   1.70312  0.00283347   0.0302411

Total time run:         30.822350

Total writes made:      31032

Write size:             4096

Object size:            4096

Bandwidth (MB/sec):     3.93282

Stddev Bandwidth:       3.66265

Max bandwidth (MB/sec): 13.668

Min bandwidth (MB/sec): 0

Average IOPS:           1006

Stddev IOPS:            937

Max IOPS:               3499

Min IOPS:               0

Average Latency(s):     0.0317779

Stddev Latency(s):      0.164076

Max latency(s):         2.27707

Min latency(s):         0.0013848

Cleaning up (deleting benchmark objects)

Clean up completed and total clean up time :20.166559

 

 

 

On Wed, Oct 18, 2017 at 8:51 AM, Maged Mokhtar <mmokhtar@xxxxxxxxxxx> wrote:

First a general comment: local RAID will be faster than Ceph for a single threaded (queue depth=1) io operation test. A single thread Ceph client will see at best same disk speed for reads and for writes 4-6 times slower than single disk. Not to mention the latency of local disks will much better. Where Ceph shines is when you have many concurrent ios, it scales whereas RAID will decrease speed per client as you add more.

Having said that, i would recommend running rados/rbd bench-write and measure 4k iops at 1 and 32 threads to get a better idea of how your cluster performs:

ceph osd pool create testpool 256 256
rados bench -p testpool -b 4096 30 write -t 1
rados bench -p testpool -b 4096 30 write -t 32
ceph osd pool delete testpool testpool --yes-i-really-really-mean-it

rbd bench-write test-image --io-threads=1 --io-size 4096 --io-pattern rand --rbd_cache=false
rbd bench-write test-image --io-threads=32 --io-size 4096 --io-pattern rand --rbd_cache=false

I think the request size difference you see is due to the io scheduler in the case of local disks having more ios to re-group so has a better chance in generating larger requests. Depending on your kernel, the io scheduler may be different for rbd (blq-mq) vs sdx (cfq) but again i would think the request size is a result not a cause.

Maged

On 2017-10-17 23:12, Russell Glaue wrote:

I am running ceph jewel on 5 nodes with SSD OSDs.

I have an LVM image on a local RAID of spinning disks.

I have an RBD image on in a pool of SSD disks.

Both disks are used to run an almost identical CentOS 7 system.

Both systems were installed with the same kickstart, though the disk partitioning is different.

 

I want to make writes on the the ceph image faster. For example, lots of writes to MySQL (via MySQL replication) on a ceph SSD image are about 10x slower than on a spindle RAID disk image. The MySQL server on ceph rbd image has a hard time keeping up in replication.

 

So I wanted to test writes on these two systems

I have a 10GB compressed (gzip) file on both servers.

I simply gunzip the file on both systems, while running iostat.

 

The primary difference I see in the results is the average size of the request to the disk.

CentOS7-lvm-raid-sata writes a lot faster to disk, and the size of the request is about 40x, but the number of writes per second is about the same

This makes me want to conclude that the smaller size of the request for CentOS7-ceph-rbd-ssd system is the cause of it being slow.

 

 

How can I make the size of the request larger for ceph rbd images, so I can increase the write throughput?

Would this be related to having jumbo packets enabled in my ceph storage network?

 

 

Here is a sample of the results:

 

[CentOS7-lvm-raid-sata]

$ gunzip large10gFile.gz &

$ iostat -x vg_root-lv_var -d 5 -m -N

Device:         rrqm/s   wrqm/s     r/s     w/s    rMB/s    wMB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util

...

vg_root-lv_var     0.00     0.00   30.60  452.20    13.60   222.15  1000.04     8.69   14.05    0.99   14.93   2.07 100.04

vg_root-lv_var     0.00     0.00   88.20  182.00    39.20    89.43   974.95     4.65    9.82    0.99   14.10   3.70 100.00

vg_root-lv_var     0.00     0.00   75.45  278.24    33.53   136.70   985.73     4.36   33.26    1.34   41.91   0.59  20.84

vg_root-lv_var     0.00     0.00  111.60  181.80    49.60    89.34   969.84     2.60    8.87    0.81   13.81   0.13   3.90

vg_root-lv_var     0.00     0.00   68.40  109.60    30.40    53.63   966.87     1.51    8.46    0.84   13.22   0.80  14.16

...

 

[CentOS7-ceph-rbd-ssd]

$ gunzip large10gFile.gz &

$ iostat -x vg_root-lv_data -d 5 -m -N

Device:         rrqm/s   wrqm/s     r/s     w/s    rMB/s    wMB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util

...

vg_root-lv_data     0.00     0.00   46.40  167.80     0.88     1.46    22.36     1.23    5.66    2.47    6.54   4.52  96.82

vg_root-lv_data     0.00     0.00   16.60   55.20     0.36     0.14    14.44     0.99   13.91    9.12   15.36  13.71  98.46

vg_root-lv_data     0.00     0.00   69.00  173.80     1.34     1.32    22.48     1.25    5.19    3.77    5.75   3.94  95.68

vg_root-lv_data     0.00     0.00   74.40  293.40     1.37     1.47    15.83     1.22    3.31    2.06    3.63   2.54  93.26

vg_root-lv_data     0.00     0.00   90.80  359.00     1.96     3.41    24.45     1.63    3.63    1.94    4.05   2.10  94.38

...

 

[iostat key]

w/s == The number (after merges) of write requests completed per second for the device.

wMB/s == The number of sectors (kilobytes, megabytes) written to the device per second.

avgrq-sz == The average size (in kilobytes) of the requests that were issued to the device.

avgqu-sz == The average queue length of the requests that were issued to the device.

 

 

_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

 

 

 

 

 

 

 

 

_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com