lan card speed very slow

[Salm]

HI

I'm working on rx2620 Itanium servers and using 4 lan cards and USER, BACKUP,DATA and MANAGEMENT.

I got a problem . backup guy complaining NIC card for backup is very slow . they are not able to backup.I kkep the setting as network team adive which is 100 HALF DUPLEX AUTONEG ON.

this utility is on CORE lan card ,

Any body can gicve me the reason ? why Backup is very slow ??

[Torsten.]

This is probably related to your other problem:

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=1164833

Is this already resolved?

[Pete Randall]

The 100Base/T cards work best when both ends of the connection are set to 100FD. They simply do not auto-negotiate well.

You can use the landamin command to set the speed on the NIC: landmin -x 100FD 0 (where 0 is the ppa number from lanscan). You then need to change the appropriate hp*conf file based on the driver used for this card:

btlan hpbtlanconf
btlan0 hpeisabtconf
btlan1 hpbasetconf
btlan3 hpbase100conf
btlan4 hpgsc100conf
btlan5 hppci100conf
btlan6 hpsppci100conf


Pete

[Tim Nelson]

Half duplex is like an overcrowded concert with only one door in and out.

You really need to talk with your network admins, not since the 90s have things run at HD..

Set all ports to 100FD with lanadmin and tell network admins to set the same at the switch.

[Torsten.]

And BTW, the core I/O card on this server is able to do 1000!!!

[Salm]

Hi
Thanks for sendig me the ling . unfortunaltey I"m not able to open this link
I'm trying for long time

Can u advice what would be the other issues ??

[Torsten.]

The link is your previous thread named "problem in /etc/hosts file" - solved?

[Tim Nelson]

As Torsten mentioned, the core cards are 1000MB.

These should be set to auto at the server and switch end.

[Jim Keeble · Accepted solution]

As every one has said, their is a mismatch in the duplex setting between the NIC and the switch port. Setting the NIC to auto-negotiate and the switch to 100FD is NOT ok. If one side is auto-negotiating, and the other is hard-coded, the auto-negotiating side MUST default to half duplex.

Why is it slow? The side that defaults to half duplex uses carrier sense/collision detection to know if it is "stepping on" someone else's transmission, or is being "stepped on". The side that is set to full duplex, on the other hand, thinks it is free to send packets whenever it wants, which is does by ignoring collision detection.

The auto/half side will be logging a lot of collision errors (Verify with lanadmin -g X, where lanX is the interface name). If the network admin checks the switch port stats, they will see lots of FCS or alignment type errors (No collisions, because the switch port has collision detect disabled). The result is a ton of packets being retransmitted unnecessarily, and the transfer runs slowly.

[rick jones]

It would be _late_ collisions which are an indication of a duplex mismatch. Attached is my boilerplate on autoneg. I am of the camp which believes that everything should be autoneg these days unless one has a hopelessly broken switch which cannot be replaced.

[Salm]

Hi I Atteched someout put

using lan1 as utility network

[rick jones]

The second half of the lanadmin -g mibstats 1 output has been omitted. While I suspect it would all be zeros since there are zeros for inbound and outbound errors, still would be best to know for certain.