[PVE-User] High ceph OSD latency

Fabrizio Cuseo f.cuseo at panservice.it
Thu Jan 15 14:28:36 CET 2015


Hi Lindsay.

First node:

root at nodo01:~# pveperf /var/lib/ceph/osd/ceph-0
CPU BOGOMIPS:      40000.76
REGEX/SECOND:      954062
HD SIZE:           1857.11 GB (/dev/sdb1)
BUFFERED READS:    152.38 MB/sec
AVERAGE SEEK TIME: 15.07 ms
FSYNCS/SECOND:     48.11
DNS EXT:           51.79 ms
DNS INT:           62.76 ms (domain.test2)

root at nodo01:~# pveperf /var/lib/ceph/osd/ceph-1
CPU BOGOMIPS:      40000.76
REGEX/SECOND:      972176
HD SIZE:           1857.11 GB (/dev/sdd1)
BUFFERED READS:    141.72 MB/sec
AVERAGE SEEK TIME: 18.91 ms
FSYNCS/SECOND:     41.38
DNS EXT:           23.32 ms
DNS INT:           79.97 ms (domain.test2)

root at nodo01:~# pveperf /var/lib/ceph/osd/ceph-2
CPU BOGOMIPS:      40000.76
REGEX/SECOND:      956704
HD SIZE:           1857.11 GB (/dev/sde1)
BUFFERED READS:    157.24 MB/sec
AVERAGE SEEK TIME: 14.97 ms
FSYNCS/SECOND:     43.48
DNS EXT:           20.50 ms
DNS INT:           130.27 ms (domain.test2)

--------
Second node:

root at nodo02:~# pveperf  /var/lib/ceph/osd/ceph-3
CPU BOGOMIPS:      39999.04
REGEX/SECOND:      965952
HD SIZE:           1857.11 GB (/dev/sdb1)
BUFFERED READS:    147.61 MB/sec
AVERAGE SEEK TIME: 22.60 ms
FSYNCS/SECOND:     42.29
DNS EXT:           45.84 ms
DNS INT:           54.82 ms (futek.it)

root at nodo02:~# pveperf  /var/lib/ceph/osd/ceph-4
CPU BOGOMIPS:      39999.04
REGEX/SECOND:      956254
HD SIZE:           1857.11 GB (/dev/sdc1)
BUFFERED READS:    143.70 MB/sec
AVERAGE SEEK TIME: 15.33 ms
FSYNCS/SECOND:     47.33
DNS EXT:           20.91 ms
DNS INT:           20.76 ms (futek.it)

root at nodo02:~# pveperf  /var/lib/ceph/osd/ceph-5
CPU BOGOMIPS:      39999.04
REGEX/SECOND:      996038
HD SIZE:           1857.11 GB (/dev/sdd1)
BUFFERED READS:    150.55 MB/sec
AVERAGE SEEK TIME: 15.83 ms
FSYNCS/SECOND:     52.12
DNS EXT:           20.69 ms
DNS INT:           21.33 ms (futek.it)

--------
Third node:

root at nodo03:~# pveperf /var/lib/ceph/osd/ceph-6
CPU BOGOMIPS:      40001.56
REGEX/SECOND:      988544
HD SIZE:           1857.11 GB (/dev/sdb1)
BUFFERED READS:    125.93 MB/sec
AVERAGE SEEK TIME: 18.15 ms
FSYNCS/SECOND:     43.85
DNS EXT:           40.32 ms
DNS INT:           22.03 ms (futek.it)

root at nodo03:~# pveperf /var/lib/ceph/osd/ceph-7
CPU BOGOMIPS:      40001.56
REGEX/SECOND:      963925
HD SIZE:           1857.11 GB (/dev/sdc1)
BUFFERED READS:    111.99 MB/sec
AVERAGE SEEK TIME: 18.33 ms
FSYNCS/SECOND:     26.52
DNS EXT:           26.22 ms
DNS INT:           20.57 ms (futek.it)

root at nodo03:~# pveperf /var/lib/ceph/osd/ceph-8
CPU BOGOMIPS:      40001.56
REGEX/SECOND:      998566
HD SIZE:           1857.11 GB (/dev/sdd1)
BUFFERED READS:    149.53 MB/sec
AVERAGE SEEK TIME: 14.75 ms
FSYNCS/SECOND:     43.25
DNS EXT:           15.37 ms
DNS INT:           55.12 ms (futek.it)


I can only see that OSD ceph-7 has less (half) fsyncs/second (also testing again it).

Those servers, have this controller:
00:1f.2 SATA controller: Intel Corporation 82801IR/IO/IH (ICH9R/DO/DH) 6 port SATA Controller [AHCI mode] (rev 02)


On the other production cluster (with Dell 2950 and PERC5i, i have really better fsyncs/sec); I think that is because of controller's cache:

root at proxmox-1:~# pveperf /var/lib/ceph/osd/ceph-0
CPU BOGOMIPS:      37238.64
REGEX/SECOND:      901248
HD SIZE:           925.55 GB (/dev/sdb1)
BUFFERED READS:    101.61 MB/sec
AVERAGE SEEK TIME: 17.39 ms
FSYNCS/SECOND:     1817.31
DNS EXT:           43.65 ms
DNS INT:           2.87 ms (panservice.it)

02:0e.0 RAID bus controller: Dell PowerEdge Expandable RAID controller 5



On the thest cluster, with only one 500Gbyte HD and this controller:

00:1f.2 IDE interface: Intel Corporation NM10/ICH7 Family SATA Controller [IDE mode] (rev 01)

i have this result:

root at nodo1:~# pveperf /var/lib/ceph/osd/ceph-1
CPU BOGOMIPS:      8532.80
REGEX/SECOND:      818860
HD SIZE:           460.54 GB (/dev/sdb1)
BUFFERED READS:    71.33 MB/sec
AVERAGE SEEK TIME: 22.75 ms
FSYNCS/SECOND:     53.49
DNS EXT:           69.40 ms
DNS INT:           2.34 ms (panservice.it)

So, for the first and third cluster, i have similar fsync/second result, but very different delay on OSDs.

I'll try to investigate some controller-related issue.

Thanks, Fabrizio





----- Messaggio originale -----
Da: "Lindsay Mathieson" <lindsay.mathieson at gmail.com>
A: pve-user at pve.proxmox.com, "Fabrizio Cuseo" <f.cuseo at panservice.it>
Inviato: Giovedì, 15 gennaio 2015 13:17:07
Oggetto: Re: [PVE-User] High ceph OSD latency

On Thu, 15 Jan 2015 11:25:44 AM Fabrizio Cuseo wrote:
> What is strange is that on OSD tree I have high latency: tipically Apply
> latency is between 5 and 25, but commit lattency is between 150 and 300
> (and sometimes 5/600), with 5/10 op/s and some B/s rd/wr (i have only 3
> vms, and only 1 is working now, so the cluster is really unloaded).
> 
> I am using a pool with 3 copies, and I have increased pg_num to 256 (the
> default value of 64 is too low); but OSD latency is the same with a
> different pg_num value.
> 
> I have other clusters (similar configuration, using dell 2950, dual ethernet
> for ceph and proxmox, 4 x OSD with 1Tbyte drive, perc 5i controller), with
> several vlms, and the commit and apply latency is 1/2ms.
> 
> Another cluster (test cluster) with 3 x dell PE860, with only 1 OSD per
> node, have better latency (10/20 ms).
> 
> What can i check ? 


POOMA U, but if you have one drive or controller that is marginal or failing, 
it can slow down the whole cluster.

Might be worth while benching individual osd's

-- 
---
Fabrizio Cuseo - mailto:f.cuseo at panservice.it
Direzione Generale - Panservice InterNetWorking
Servizi Professionali per Internet ed il Networking
Panservice e' associata AIIP - RIPE Local Registry
Phone: +39 0773 410020 - Fax: +39 0773 470219
http://www.panservice.it  mailto:info at panservice.it
Numero verde nazionale: 800 901492



More information about the pve-user mailing list