Why do you think the performance is slow? Compared to what?
I have already mentioned here and on a ceph day, when are we (ceph
community+redhat) going to put some benchmark tests out there. So people
can compare and know what to expect. It would be also nice than in
release notes to mention the measured difference or % performance
increase, instead of just writing this is 'faster' or that is
'optimized'.
I also think it is nice for the development/test team to have some
testing clusters available so they can test more before releasing
something.
-----Original Message-----
Subject: [ceph-users] poor cephFS performance on Nautilus 14.2.9
deployed by ceph_ansible
Hi guys,
We just deployed a CEPH 14.2.9 cluster with the following hardware:
MDSS x 1
Xeon Gold 5122 3.6Ghz
192GB
Mellanox ConnectX-4 Lx 25GbE
MON x 3
Xeon Bronze 3103 1.7Ghz
48GB
Mellanox ConnectX-4 Lx 25GbE
6 x 600GB 10K SAS
OSD x 5
Xeon Silver 4110 2.1Ghz x 2
192GB
Mellanox ConnectX-4 Lx 25GbE
16 x 10TB 7.2K NLSAS (block)
2 x 2TB Intel P4600 NVMe (block.db)
Network is all Mellanox SN2410/SN2700 configured at 25GbE for both front
and back network.
Just for POC at this stage, the cluster was deployed by ceph_ansible
without much customization and the initial test on its cephFS FUSE mount
performance seems to be very low. We did some test with iozone the
result as follow:
]# /opt/iozone/bin/iozone -i 0 -i 1-r 128k -s 5G -t 20
Iozone: Performance Test of File I/O
Version $Revision: 3.465 $
Compiled for 64 bit mode.
Build: linux-AMD64
Contributors:William Norcott, Don Capps, Isom Crawford, Kirby
Collins
Al Slater, Scott Rhine, Mike Wisner, Ken Goss
Steve Landherr, Brad Smith, Mark Kelly, Dr. Alain
CYR,
Randy Dunlap, Mark Montague, Dan Million, Gavin
Brebner,
Jean-Marc Zucconi, Jeff Blomberg, Benny Halevy,
Dave Boone,
Erik Habbinga, Kris Strecker, Walter Wong, Joshua
Root,
Fabrice Bacchella, Zhenghua Xue, Qin Li, Darren
Sawyer,
Vangel Bojaxhi, Ben England, Vikentsi Lapa,
Alexey Skidanov.
Run began: Tue Jun 2 16:40:53 2020
File size set to 5242880 kB
Command line used: /opt/iozone/bin/iozone -i 0 -i 1-r -s 5G -t
20 128k
Output is in kBytes/sec
Time Resolution = 0.000001 seconds.
Processor cache size set to 1024 kBytes.
Processor cache line size set to 32 bytes.
File stride size set to 17 * record size.
Throughput test with 20 processes
Each process writes a 5242880 kByte file in 4 kByte records
Children see throughput for 20 initial writers = 35001.12
kB/sec
Parent sees throughput for 20 initial writers = 34967.65
kB/sec
Min throughput per process = 1748.22
kB/sec
Max throughput per process = 1751.62
kB/sec
Avg throughput per process = 1750.06
kB/sec
Min xfer = 5232724.00 kB
Children see throughput for 20 rewriters = 35704.79
kB/sec
Parent sees throughput for 20 rewriters = 35704.30
kB/sec
Min throughput per process = 1783.44
kB/sec
Max throughput per process = 1786.29
kB/sec
Avg throughput per process = 1785.24
kB/sec
Min xfer = 5234532.00 kB
Children see throughput for 20 readers = 49368539.50
kB/sec
Parent sees throughput for 20 readers = 49317231.38
kB/sec
Min throughput per process = 2414424.00
kB/sec
Max throughput per process = 2599996.25
kB/sec
Avg throughput per process = 2468426.98
kB/sec
Min xfer = 4868708.00 kB
Children see throughput for 20 re-readers = 48675891.50
kB/sec
Parent sees throughput for 20 re-readers = 48617335.67
kB/sec
Min throughput per process = 2316395.25
kB/sec
Max throughput per process = 2703868.75
kB/sec
Avg throughput per process = 2433794.58
kB/sec
Min xfer = 4491704.00 kB
We also did some dd tests, the write speed on a single test on our
standard server is ~50MB/s but on a very big memory server, the speed is
double ~ 80-90MB/s.
We have zero experience on ceph and as said we haven't done more tuning
at this stage. But if this sort of performance is way too low from those
hardware spec?
Any hints will be appreciated.
Cheers
D
_______________________________________________
ceph-users mailing list -- [email protected] To unsubscribe send an
email to [email protected]
_______________________________________________
ceph-users mailing list -- [email protected]
To unsubscribe send an email to [email protected]
