LinuxQuestions.org - Slow File Transfers (LAN)

- Slackware (https://www.linuxquestions.org/questions/slackware-14/)

- - Slow File Transfers (LAN) (https://www.linuxquestions.org/questions/slackware-14/slow-file-transfers-lan-905774/)

Slow File Transfers (LAN)

I been having issues where I start a transfer over nfs or cifs and it starts ok at about 120MBps over the LAN, than the transfer speeds starts decreasing all the way down to like 1KBps a few seconds pass and it bursts back to 50MBps and it keeps fluctuating the same way with different speeds till its done... Its very annoying!
a simple 700MB iso takes me like 5 min to transfer. Please not that is not a network issue since I can transfer files with my macs just fine.
This issue is isolated to the slackware systems only.
I have two Slackware 13.37 x86_64 systems that do the same thing.

LAN is a GigE network at 1GBps

Any ideas?

TIA!

Check your network device for LAN throughput and make sure the device does not have any preset management tools configured to limit speeds and bandwidth based on usage if the usage gets heavy.

If you do, set the port your workstation and/or the server to EXEMPT or similar control status. Often this is labeled as exempt, disabled, or maximum bandwidth more most switches and routers.

Also, make sure you have the appropriate cabling for your network. Make sure you are using at minimum CAT5e Ethernet rated cables. Avoid using Hubs devices as well, and use Switches instead.

Slackware shouldn't be the problem in this case, and neither should the workstation or the servers.

Quote:

Originally Posted by ReaperX7 (Post 4486519)

Sorry but you are wrong. Slackware is the problem in this case. The LAN or the firewalls are not enforcing QoS or Traffic Shaping. I can not make nothing out of this one. Its only this two machines... I can take them too different areas of the house and they do the same thing.

And to my test it seems at the protocol level:

Client connecting to 10.30.2.52, TCP port 5001
TCP window size: 16.0 KByte (default)
------------------------------------------------------------
[ 3] local 10.30.2.51 port 50011 connected with 10.30.2.52 port 5001
[ ID] Interval Transfer Bandwidth
[ 3] 0.0-10.0 sec 1.10 GBytes 942 Mbits/sec
root@oscuridad:/media#

Quote:

Originally Posted by SeRi@lDiE (Post 4486406)

Any ideas?

TIA!

--edit--
duplicate info that SeRi@lDiE posted at the same time :)

Well you posted when I did. And iperf looks fine. Looks as if the hardware and wiring is just fine.

What do your smb.conf / /etc/exports and mount options look like?

Also, could you run iperf with a longer test, just to make sure there isn't a buffer problem on the NICs. --time 120 instead of the default 10 seconds?

Quote:

Originally Posted by disturbed1 (Post 4486560)

disturbed1,

Thanks for the help. I will post the results in a bit. I have my hands tide with a RHEL5 install. :)

disturbed1,

Results:

Code:

root@oscuridad:/media# iperf -c 10.30.2.52 --time 120

------------------------------------------------------------

Client connecting to 10.30.2.52, TCP port 5001

TCP window size: 16.0 KByte (default)

------------------------------------------------------------

[  3] local 10.30.2.51 port 47956 connected with 10.30.2.52 port 5001

[ ID] Interval      Transfer    Bandwidth

[  3]  0.0-120.0 sec  13.2 GBytes  941 Mbits/sec

root@oscuridad:/media#

Pretty consistent if you ask me :)

Code:

nfs:



root@externo:~# cat /etc/exports 

# See exports(5) for a description.

# This file contains a list of all directories exported to other computers.

# It is used by rpc.nfsd and rpc.mountd.

/mnt/backup/    10.30.2.0/24(rw,no_root_squash)

root@externo:~# 



cifs:



[ global ]

interfaces = egiga0

unix charset = UTF8

workgroup = ROOT

netbios name = nas-node

server string = DNS-323

hosts allow = 

hosts deny = 

security = USER

encrypt passwords = yes

max log size = 0

socket options = TCP_NODELAY SO_RCVBUF=65536 SO_SNDBUF=65536

max xmit = 65535

create mask = 0777

directory mask  = 0777

force create mode = 0777

force directory mode = 0777

load printers = Yes

printcap name = /usr/local/LPRng/etc/printcap

min print space = 2000

max print jobs = 1000

printing = lprng

print command = /usr/local/LPRng/bin/lpr -P%p -r %s 

lpq command = /usr/local/LPRng/bin/lpq -P%p

lprm command = /usr/local/LPRng/bin/lprm -P%p %j

lppause command = /usr/local/LPRng/sbin/lpc hold %p %j

lpresume command = /usr/local/LPRng/sbin/lpc release %p %j

queuepause command = /usr/local/LPRng/sbin/lpc -P%p stop

queueresume command = /usr/local/LPRng/sbin/lpc -P%p start

use sendfile =yes





[ web_page ]

comment = Enter Our Web Page Setting

path = /mnt/web_page

valid users = 

read only = yes

guest ok = yes



[printers]

path = /mnt/HD_a2/.lpd

guest ok = Yes

printable = Yes

use client driver = Yes

browseable = No



[ BT ]

comment = 

path = /mnt/HD_a2/BT

valid users = 

read only = no

guest ok = yes

oplocks = no

map archive = no



[ Volume_1 ]

comment = 

path = /mnt/HD_a2

valid users = nUll

read only = no

guest ok = no

oplocks = no

map archive = no



root@nas-node:~#

Like the iperf. To me appears to rule any questions on the hardware and wires. Some Intel NICS have known eeprom issues, that are never seen with short bursts of traffic. I'm sure Intel is not the only NIC that has a similar issue.

Considering I don't have any transfer issues, I'll point out some of the differences I see.

My smb.conf does not have any socket options. I do recall reading different docs that stated different send and receive buffers to tune performance. Changing these values never yielded better performance, and in some cases hurt performance.

Quote:

SO_SNDBUF and SO_RCVBUF
The send and receive buffers can often be the reset to a value higher than that of the operating system. This yields a marginal increase of speed (until it reaches a point of diminishing returns).

http://oreilly.com/openbook/samba/book/appb_02.html

You can modify /etc/rc.d/rc.nfs in an attempt to squeeze more performance out of NFS.

This is my setting

Code:

# Start 8 nfsd servers by default (an old Sun standard):

  if [ -x /usr/sbin/rpc.nfsd ]; then

    echo "  /usr/sbin/rpc.nfsd 8"

    /usr/sbin/rpc.nfsd -U 32

  fi

I launch 32 NFS servers, -U specifies TCP instead of UDP.

copy ~1400M file from disturbed1 to backend

Code:

keith@disturbed1:~$ mount | grep Share

backend:/mnt/Share on /mnt/Share type nfs (rw,noatime,intr,tcp,ac,addr=192.168.1.111)

//backend/Share on /mnt/tmp type cifs (rw,mand)

keith@disturbed1:~$ ls -la test.bin 

-rw-r--r-- 1 keith users 1468006400 Sep 30 14:49 test.bin



keith@disturbed1:~$ scp test.bin keith@backend:/mnt/Share/

keith@backend's password: 

test.bin                                                                                                                            100% 1400MB  66.7MB/s  00:21    



NFS

keith@disturbed1:~$ dd if=/mnt/Share/test.bin of=/dev/null 

2867200+0 records in

2867200+0 records out

1468006400 bytes (1.5 GB) copied, 18.0938 s, 81.1 MB/s



CIFS

keith@disturbed1:~$ dd if=/mnt/tmp/test.bin of=/dev/null 

2867200+0 records in

2867200+0 records out

1468006400 bytes (1.5 GB) copied, 1.44791 s, 1.0 GB/s <---- :D file caching 



( a few syncs, flushes, mount/unmounts)

keith@disturbed1:~$ dd if=/mnt/tmp/test.bin of=/mnt/media/test.bin

2867200+0 records in

2867200+0 records out

1468006400 bytes (1.5 GB) copied, 13.3679 s, 110 MB/s

----
System info -
disturbed1 = 1090t x6 Slackware64 -current with Kernel 3.0.3 RTL 8168 onboard nic.
backend = e8500 core2duo Slackware64 13.37 with Kernel 2.6.38.4 RTL 8168 onboard nic.
7200 SATA drives.

You said QoS isn't working eh...

Have you tried AlienBOB's Easy Firewall Generator to setup for QoS packet scheduling via IPTables/IPChains?

There should also be an administration tool if you use webmin.

I also found this:

Code:

http://www.lartc.org/howto/

Are you trying to write to an NTFS file system when you do the transfers?

Quote:

root@nas-node:~#

suggests you are attempting to write to some NAS device
I get the impression that the LAN configuration is all OK. The problem comes with writing the transferred data. I have seen slowing write performance when writing large files to NTFS file systems.

Quote:

Originally Posted by allend (Post 4486834)

Are you trying to write to an NTFS file system when you do the transfers?
suggests you are attempting to write to some NAS device
I get the impression that the LAN configuration is all OK. The problem comes with writing the transferred data. I have seen slowing write performance when writing large files to NTFS file systems.

Yes one is a NAS and two are slackware. All systems are Ext3.

Quote:

Originally Posted by ReaperX7 (Post 4486793)

Code:

http://www.lartc.org/howto/

What? Please stay on topic. I never said that they are not working nether I want them to work. I was stating that They are not "set".

Thanks for the help though.

Quote:

Originally Posted by SeRi@lDiE (Post 4486558)

The LAN or the firewalls are not enforcing QoS or Traffic Shaping.

What makes you so sure? The fact that your Macs are doing a better job is not proof.

Quote:

Originally Posted by SeRi@lDiE (Post 4486406)

Are you writing to a D-Link dns323 nas enclosure?
If so then both reads and writes to that thing are somewhere between 8-20 Mb/s.
On my Gb lan I can't get more than about 12Mb/s speed out of the dns323 with transfers from/to both Slackware and Win7 systems.