LinuxQuestions.org
Welcome to the most active Linux Forum on the web.
Home Forums Tutorials Articles Register
Go Back   LinuxQuestions.org > Forums > Linux Forums > Linux - Networking
User Name
Password
Linux - Networking This forum is for any issue related to networks or networking.
Routing, network cards, OSI, etc. Anything is fair game.

Notices


Reply
  Search this Thread
Old 05-31-2010, 12:43 AM   #1
dpnctl
LQ Newbie
 
Registered: Mar 2010
Posts: 15

Rep: Reputation: 0
tuning RHEL for "long fat" links


Hi,
I have an application pushing data on long WAN links.
The application runs on RHEL4.
It is not able to utilise the complete available bandwidth.
I was told that kernel needs to be tuned for "long fat" links, so that the system starts utilising complete available bandwidth.
I found a link "http://www.kernelfaq.com/2007/07/tcp-tuning-sysctls.html", however it is on FreeBSD.
Does any one have info on this.
Regards
Mahen
 
Old 05-31-2010, 05:30 PM   #2
lazlow
Senior Member
 
Registered: Jan 2006
Posts: 4,363

Rep: Reputation: 172Reputation: 172
Are you running GigE or fast Ethernet? How fast are you actually moving data? With fast Ethernet anything over about 9.5MB/sec is probably the best you are going to get without a ton of work. You can use channel bonding but by the time you get all the extra equipment and get it working you are usually better off(and cheaper) just moving to GigE. On GigE you should be seeing 95MB/sec(assuming your hard drives can handle that fast).
 
Old 06-01-2010, 11:53 AM   #3
dpnctl
LQ Newbie
 
Registered: Mar 2010
Posts: 15

Original Poster
Rep: Reputation: 0
Smile tuning TCP for "long fat" links

Hi ,
I am sorry for not giving complete info.
The OS is RHEL 4 with 2.6.9 kernel. The auto tuning is set to on.

The application runs on a Dell 2950 with GigE network, however bonding is not supported by this app.

The available pipe is 135 MBps, unfortunately the application is using only 6 Mbps, ealier.

www.psc.edu/networking/projects/tcptune/#Linux - is the link we followed.

Today our manager has made some changes in the propreietry application and we have seen a good performance change.

We changed the default tcp_wmem on source and tcp_rmem on destination, tested the params with iperf (transmitting 1 MB packets).

now its going till 2MBps :-), i believe with your suggestions we can get more.



Any insight on this? do we need additional info?

mahen
 
Old 06-01-2010, 05:30 PM   #4
lazlow
Senior Member
 
Registered: Jan 2006
Posts: 4,363

Rep: Reputation: 172Reputation: 172
dpnctl

Just to be sure. Mbps is the standard for transmission rates(Meba bit per second).Mbps does not equal MBps(factor of eight difference). For instance fast ethernet is 100Mbps, not 100MBps(Mega Byte per second).

My guess is that your pipe is actually 135Mbps.

Bonding would be at the OS level not at the application, which is irrelevant since you are on GigE.

For the transmit levels to be that low you almost have to be defaulting to fast Ethernet or even base 10. In order for GigE to work the entire path the data follow HAS to be GigE equipment. Every cable, switch, router, and machine that the data goes through has to be GigE. Start by checking with ethtool on both ends to make sure they are actually running GigE.

Quote:
[root@localhost ~]# ethtool eth0
Settings for eth0:
Supported ports: [ MII ]
Supported link modes: 10baseT/Half 10baseT/Full
100baseT/Half 100baseT/Full
1000baseT/Full
Supports auto-negotiation: Yes
Advertised link modes: 10baseT/Half 10baseT/Full
100baseT/Half 100baseT/Full
1000baseT/Full
Advertised auto-negotiation: Yes
Speed: 100Mb/s
Duplex: Full
Port: MII
PHYAD: 9
Transceiver: external
Auto-negotiation: on
Supports Wake-on: g
Wake-on: d
Link detected: yes
Note in this example I have a GigE card but I am hooked up to a Fast Ethernet device. Therefore it drops back to the fast Ethernet speed.

Now for the bad news, all that tells us about is the connection between the machine and the first device it connects to (switch, router, machine, etc). Even if I had been running at GigE on this first hop, the next hop might only be fast Ethernet. The only way I could know this (from my machine) was by the fact that my throughput was low(like yours is). Even if every hop on the path is GigE equipment you could have a failure anywhere along the line. Cables (particularly with homemade ends) are notorious for this. The other thing to look out for is if somebody ran the cable too long. It may only be 100meters(GigE limit) from point A to point B but that is irrelevant, it is the actual length of the cable being used. IT guys are notorious for using two 75 meter cables with a (passive) connector and then wonder why they are having intermittent issues.

Just a few things to check off the top of my head.
 
  


Reply



Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is Off
HTML code is Off



Similar Threads
Thread Thread Starter Forum Replies Last Post
RHEL install Vostro 400 - sys "Loading SCSI drivers" "loading ata_piix driver..." sreekr_news Linux - Newbie 11 01-21-2010 01:25 PM
How to convert "unsigned long long" to "char pointer"? novicehacker Linux - Kernel 1 11-20-2006 12:57 AM
LONG LONG "lost" /home directory post Mark_in_Hollywood Linux - Newbie 5 09-19-2006 11:47 PM
Lost DHCP Lease when putting computer in "Standby" mode for a "long" time pnellesen Linux - Networking 1 01-06-2005 11:44 PM
Kernel Error " kern.maxpipekva exceeded, please see tuning(7)." nazzymac *BSD 3 11-17-2004 05:29 PM

LinuxQuestions.org > Forums > Linux Forums > Linux - Networking

All times are GMT -5. The time now is 03:53 PM.

Main Menu
Advertisement
My LQ
Write for LQ
LinuxQuestions.org is looking for people interested in writing Editorials, Articles, Reviews, and more. If you'd like to contribute content, let us know.
Main Menu
Syndicate
RSS1  Latest Threads
RSS1  LQ News
Twitter: @linuxquestions
Open Source Consulting | Domain Registration