Network performance - SAMBA - NFS

All,

I'd like to gather the various threads that reported Samba and/or NFS performance benchmarks for the latest kernels (3.16 or newer). I want to track these comparing to stock OS on various Kirkwood boxes.

Please provide the links to such posts in this forum that you are aware of.

-bodhi
===========================
Forum Wiki
bodhi's corner (buy bodhi a beer)

As an addendum to bodhi's request, If you want to perform the test yourself and would like a standard way of doing this so we can all compare and have a baseline set to work with. Ideally you should be connected at Gbit speeds on a Gbit switch - not on wireless or 100Mbit switch.

Do the following
1)..Start with a massive and compressed file...This should be a close to 500mb file.
2)..Then find you architecture and kernel version
3)..Then mount your samba/nfs share and copy the file to and from your device (PUSH - PULL)


Create the file - called "bigfile"

dd if=/dev/urandom of=bigfile bs=512 count=1000000

Find your architecture and kernel details

uname -ar

PUSH the file to the location

time cp bigfile /mnt/samba/server/share #where /mnt/samba/server/share = your mounted share

time cp bigfile /mnt/nfs/server/share #where /mnt/nfs/server/share = your mounted share

The "time" command will tell you how long it took.

PULL the file from the location

time cp /mnt/samba/server/share/bigfile .       #where /mnt/samba/server/share = your mounted share

time cp /mnt/nfs/server/share/bigfile .              #where /mnt/nfs/server/share = your mounted share

The "time" command will tell you how long it took.

Post the results you get, bonus points if you do it like this post 6 in this thread



Edited 5 time(s). Last edit at 07/02/2016 02:34PM by Gravelrash.

And do the same but reverse the direction (pull):

time cp /mnt/samba/server/share/bigfile .       #where /mnt/samba/server/share = your mounted share
time cp /mnt/nfs/server/share/bigfile .              #where /mnt/nfs/server/share = your mounted share

Lastly, these 4 tests should be performed in a Gigabit network, with only a Gigabit switch in the middle of the 2 computers.

-bodhi
===========================
Forum Wiki
bodhi's corner (buy bodhi a beer)



Edited 1 time(s). Last edit at 06/29/2016 11:47PM by bodhi.

May I suggest to put these together in a script, which reports the times and the samba configuration (and other stuff which might be of interest)? A standardized output would make the collection of the data much easier.

Rainer

===========================
Blog: https://rmkrug.wordpress.com/

1 x Pogoplug E02 (pink): Logitech media server

1 x Pogoplug E02 (pink): FTP server only (throw away configuration - quick to rebuild)

1 x GoFlex Net: running OpenMediaVault

Here are the results from the

"SMB - on GoFlex Net - running OpenMediaVault - share on usb flash driver - client Mac OS X"

jury:

01:01:05 ~$ dd if=/dev/urandom of=bigfile bs=512 count=1000000
1000000+0 records in
1000000+0 records out
512000000 bytes transferred in 42.935120 secs (11924970 bytes/sec)
01:07:31 ~$ time cp bigfile /Volumes/Scans/

real	2m7.070s
user	0m0.004s
sys	0m0.607s
01:09:42 ~$ time cp /Volumes/Scans/bigfile ./bigfile.down

real	0m46.476s
user	0m0.005s
sys	0m1.198s
01:12:58 ~$

Architecture and Version of server:

root@lms:~# uname -ar
Linux lms 3.2.0-4-kirkwood #1 Debian 3.2.78-1 armv5tel GNU/Linux

Here is the config from the samba server. Stock settings, no tweaks:

#======================= Global Settings =======================
[global]
workgroup = WORKGROUP
server string = %h server
dns proxy = no
log level = 0
syslog = 0
log file = /var/log/samba/log.%m
max log size = 1000
syslog only = yes
panic action = /usr/share/samba/panic-action %d
encrypt passwords = true
passdb backend = tdbsam
obey pam restrictions = yes
unix password sync = no
passwd program = /usr/bin/passwd %u
passwd chat = *Enter\snew\s*\spassword:* %n\n *Retype\snew\s*\spassword:* %n\n *password\supdated\ssuccessfully* .
pam password change = yes
socket options = TCP_NODELAY IPTOS_LOWDELAY
guest account = nobody
load printers = no
disable spoolss = yes
printing = bsd
printcap name = /dev/null
unix extensions = yes
wide links = no
create mask = 0777
directory mask = 0777
map to guest = Bad User
use sendfile = yes
aio read size = 16384
aio write size = 16384
null passwords = no
local master = yes
time server = no
wins support = no

#======================= Share Definitions =======================
[Scans]
path = /media/6c76167f-2d6d-4453-b79b-c79a57e5e948/Scans/
guest ok = yes
read only = no
browseable = yes
inherit acls = yes
inherit permissions = no
ea support = no
store dos attributes = no
printable = no
create mask = 0755
force create mode = 0644
directory mask = 0755
force directory mode = 0755
hide dot files = yes
read list =
write list =

===========================
Blog: https://rmkrug.wordpress.com/

1 x Pogoplug E02 (pink): Logitech media server

1 x Pogoplug E02 (pink): FTP server only (throw away configuration - quick to rebuild)

1 x GoFlex Net: running OpenMediaVault



Edited 1 time(s). Last edit at 06/30/2016 01:20PM by rkrug.

uname -ar

Linux Smurfville 3.16.0-kirkwood-tld-3 #2 PREEMPT Sat Sep 27 02:01:41 PDT 2014 armv5tel GNU/Linux

NFS: Pogov4 (SATA HDD) <-> Lenovo IX2-DL (SATA HDD)

dd if=/dev/urandom of=bigfile bs=512 count=1000000

time cp bigfile /mnt/network/Backups/

real	0m35.817s
user	0m0.060s
sys	0m9.010s

time cp /mnt/network/Backups/bigfile .

real	0m27.336s
user	0m0.030s
sys	0m13.330s

SAMBA: Pogov4 (SATA HDD) <-> Lenovo IX2-DL (SATA HDD)

dd if=/dev/urandom of=bigfile bs=512 count=1000000

time cp bigfile /mnt/network/samba/

real	0m35.387s
user	0m0.070s
sys	0m9.160s

time cp /mnt/network/samba/bigfile .

real	0m26.752s
user	0m0.080s
sys	0m13.070s

Edited 1 time(s). Last edit at 07/02/2016 05:23AM by Gravelrash.

@Gravelrash,

Well done :)

Also, the HP Thin Client is probably a good test bench, since we can run stock OS and see if kernel 3.16+ can compare.


@rkrug,

Could you do the same tests Gravelrash did? can you run with kernel 3.16.0-kirkwood-tld-1 or newer? that would be a good test to compare the performance. And please use Garvelrash's format as shown above.

-bodhi
===========================
Forum Wiki
bodhi's corner (buy bodhi a beer)



Edited 1 time(s). Last edit at 07/02/2016 07:28AM by bodhi.

.



Edited 1 time(s). Last edit at 07/02/2016 05:32PM by Gravelrash.

uname -ar

Linux HP5325 4.6.0-kirkwood-tld-1 #1 PREEMPT Wed May 18 03:54:14 PDT 2016 armv5tel GNU/Linux

NFS: HP T5325 (USB HDD) <-> Lenovo IX2-DL (SATA HDD)

dd if=/dev/urandom of=bigfile bs=512 count=1000000

time cp bigfile /mnt/network/nfs/Backups

real	0m51.342s
user	0m0.040s
sys	0m6.800s

time cp /mnt/network/nfs/Backups/bigfile .

real	0m43.917s
user	0m0.020s
sys	0m11.450s

SAMBA: HP T5325 (USB HDD) <-> Lenovo IX2-DL (SATA HDD)

dd if=/dev/urandom of=bigfile bs=512 count=1000000

time cp bigfile /mnt/network/samba

real	0m48.287s
user	0m0.000s
sys	0m7.630s

time cp /mnt/network/samba/bigfile .

real	0m48.146s
user	0m0.030s
sys	0m9.210s

Seems slower - than the PogoV4 - but I guess thats more down to the USB HDD not being on native SATA and the protocol overhead.



Edited 1 time(s). Last edit at 07/02/2016 05:33PM by Gravelrash.

> Seems slower - than the PogoV4 - but I guess thats
> more down to the USB HDD not being on native SATA
> and the protocol overhead.

Yes, it's SATA vs USB.

-bodhi
===========================
Forum Wiki
bodhi's corner (buy bodhi a beer)

if i get chance this week - i will load my other HP thinclient with stock os and see where we go. Im down to two now.

@Gravelrash,

If you can run stock, it's also good to run iperf to see pure network performance difference, if any, between stock T5325 and Debian kernel 3.16+.

-bodhi
===========================
Forum Wiki
bodhi's corner (buy bodhi a beer)

I know this is an older topic but I figured why not post some current numbers...

NFSv3 testing

RAID1 ext4 (rw,noatime,stripe=64,data=writeback) 2x WD RED 3TB
NSA-320

Linux IQ2700i 4.11.3-kirkwood-tld-2 #1 PREEMPT Tue Jun 6 17:01:17 PDT 2017 armv5tel GNU/Linux
# cat /etc/debian_version
8.8

READ NFSv3 - 6.21GB file (Install Mac OSX El Capitan.pkg)

ethstats peak (100% CPU usage on NAS):
  egiga0:   10.54 Mb/s In   394.53 Mb/s Out -  17013.5 p/s In   32729.7 p/s Out
iStats Menu Meters (Mac Pro 2008 10.12.5):
  51.8MB/s peak - avg ~42MB/s

WRITE NFSv3 - 6.21GB file

ethstats peak (100% CPU usage on NAS load avg 9.40):
  egiga0:  260.46 Mb/s In     1.16 Mb/s Out -  21890.0 p/s In    1020.0 p/s Out
iStats Menu Meters (Mac Pro 2008 10.12.5):
  33.3MB/s peak - avg ~29MB/s

And for the 512MB "bigfile" test:

MacPro:Desktop ludacrisvp$ dd if=/dev/urandom of=bigfile bs=512 count=1000000
1000000+0 records in
1000000+0 records out
512000000 bytes transferred in 47.029512 secs (10886781 bytes/sec)

Push to NAS

MacPro:Desktop ludacrisvp$ time cp bigfile /Volumes/data/

real	0m13.919s
user	0m0.006s
sys	0m2.048s

  egiga0:  306.75 Mb/s In     1.24 Mb/s Out -  25631.3 p/s In    1125.1 p/s Out

Pull from NAS

MacPro:Desktop ludacrisvp$ time cp /Volumes/data/bigfile ./bigfile.down

real	0m10.627s
user	0m0.006s
sys	0m1.223s
  egiga0:    6.92 Mb/s In   257.74 Mb/s Out -  11142.1 p/s In   21374.0 p/s Out

SAMBA benchmark for 2 GoFlex Home boxes (1Gbs Ethernet, 1.2GHz CPU, 128MB RAM)

Linux version 6.12.6-kirkwood-tld-1
Debian 12.8

The /etc/smb.conf on each of these 2 boxes is the Debian maintainer version (unchanged, except for Shares definition).

512MB file

root@HomeMedia2:/media/HOME# dd if=/dev/urandom of=bigfile bs=512 count=1000000

(bigfile2 and bigfile are identical)

Push media2 to media1

root@HomeMedia2:/media/HOME# time cp bigfile /mnt/homemedia1/Misc/ &

real 0m26.806s
user 0m0.011s
sys 0m8.200s

~ 20 MBs = 160 Mbs

Pull media1 to media2

root@HomeMedia2:/media/HOME/Misc# time cp /mnt/homemedia1/Misc/bigfile2 .  &

real 0m42.642s
user 0m0.017s
sys 0m17.675s

Push media1 to media2

root@HomeMedia1:/media/HOME/Misc#  time cp bigfile2 /mnt/homemedia2/Misc/ &

real 0m25.306s
user 0m0.034s
sys 0m9.046s

Pull media2 to media1

root@HomeMedia1:/media/HOME/Misc# time cp /mnt/homemedia2/Misc/bigfile2 .  &

real 0m42.653s
user 0m0.009s
sys 0m17.376s

-bodhi
===========================
Forum Wiki
bodhi's corner (buy bodhi a beer)



Edited 4 time(s). Last edit at 01/06/2025 05:06PM by bodhi.

NFS server is a goflexnet (1.2GHz CPU, 128MB RAM) NFS v4 exports untuned.

# uname -rm
6.12.6-kirkwood-tld-1 armv5tel

# cat /etc/debian_version
12.8

Client 1 is a mac

mac$ dd if=/dev/urandom of=bigfile bs=512 count=1000000
1000000+0 records in
1000000+0 records out
512000000 bytes transferred in 2.103410 secs (243414265 bytes/sec)

mac$ time cp bigfile /Volumes/gfn_vol1/

real    0m28.436s
user    0m0.003s
sys     0m1.115s

mac$ time cp /Volumes/gfn_vol1/bigfile bigfile.down

real    0m21.279s
user    0m0.003s
sys     0m0.860s

mac$ rm /Volumes/gfn_vol1/bigfile

Client 2 a Dell Kace M300 (2.0GHz CPU, 2GB RAM)

m300$ uname -rm
6.12.6-kirkwood-tld-1 armv5tel
m300$ cat /etc/debian_version
12.8

m300$ dd if=/dev/urandom of=bigfile bs=512 count=1000000
1000000+0 records in
1000000+0 records out
512000000 bytes (512 MB, 488 MiB) copied, 33.3406 s, 15.4 MB/s

m300$ time cp bigfile /nfs/gfn_vol1/

real    0m38.865s
user    0m0.002s
sys     0m7.409s

m300$ time cp /nfs/gfn_vol1/bigfile bigfile.down

real    0m26.832s
user    0m0.001s
sys     0m9.741s

All hosts connected with 1Gbit ethernet. All disks SATA SSD.

eno789,

Interesting!

I thought the server RAM is the bottleneck when you pull the files from it (see my "Pull media2 to media1"). But in your M300 test, the result is reversed of what my push and pull tests.

Quote

m300$ time cp bigfile /nfs/gfn_vol1/

real 0m38.865s
user 0m0.002s
sys 0m7.409s

m300$ time cp /nfs/gfn_vol1/bigfile bigfile.down

real 0m26.832s
user 0m0.001s
sys 0m9.741s

There is something strange going on :)

SATA HDDs were used in all my benchmarks.

-bodhi
===========================
Forum Wiki
bodhi's corner (buy bodhi a beer)