[THIN] Re: OT: NIC Teaming

  • From: "Berny Stapleton" <berny.stapleton@xxxxxxxxxx>
  • To: <thin@xxxxxxxxxxxxx>
  • Date: Mon, 5 Dec 2005 16:15:10 -0000

Your problem of a lack of negotiation would mean that you were missing one or 
more of the cable pairs.
GigE requires all 8 wires (4 pairs) for operation, not just 4 wires (Two pairs  
/ TX and RX) like Fast Ethernet.
Take a look here:
<http://www.commsdesign.com/design_corner/OEG20020822S0002 ;>     That should 
give you a bit more info.
Cisco had (have?) a reputation for link issues. Have a chat to someone who uses 
Fluke equipment for link testing, a number of the Fluke devices came with 
warning pamphlets in the boxes regarding doing link testing with Cisco switches.
Essentially if my mind remembers correctly, they didn't (Or don't) send a solid 
link signal, they blink it (Switch it on and off) to save power. If anyone 
wants to correct me on this, please feel free.


From: thin-bounce@xxxxxxxxxxxxx [mailto:thin-bounce@xxxxxxxxxxxxx] On Behalf Of 
Evan Mann
Sent: 05 December 2005 14:11
To: thin@xxxxxxxxxxxxx
Subject: [THIN] Re: OT: NIC Teaming

I've experienced odd negotiation problems with my Cisco 3560's and other 
devices, but never my Dell servers on my Catalyst 6513, with are all 
auto-negotiation on the servers and the 6513.

The problem with this 1 NIC is the cable wasn't snapped into the patch panel 
all the way, and it happened to be in a way that it could only negotiate at 100 
instead of GigE, odd.
On my 3560's, I've had problems where I force 100/full or 10/full and try to 
link up to devices that are non-Cisco (such as some IAD's from my ISP) and they 
just refuse to work.  Have to set them to auto and they link instantly.  
As for my original issues, they went away the next morning during a copy test.  
Slow for about 5 minutes, then it bumped to full speed.  Something was/is 
getting on the network and flooding out.  Probably excessive broadcasts, but 
since it comes and goes, it's really hard to track down. 


From: thin-bounce@xxxxxxxxxxxxx [mailto:thin-bounce@xxxxxxxxxxxxx] On Behalf Of 
Andrew Wood
Sent: Monday, December 05, 2005 5:08 AM
To: thin@xxxxxxxxxxxxx
Subject: [THIN] Re: OT: NIC Teaming

Come across the same thing myself. I've read a number of docs that state 
setting the port speed will gain the best results, but have found in the field 
a number of servers (dells especially) seem to operate better if the port speed 
is set to auto on the server and fixed at the switch port. 
I think ultimately this setting is depenedent on both the server and switch 
hardware, and the 'fix everything at 100mb/fd' is not a networking silver 
bullet. Perhaps a better recommendation would be to test fixing settings on 
servers and switches and implement whatever comes out best in your environment?

From: thin-bounce@xxxxxxxxxxxxx [mailto:thin-bounce@xxxxxxxxxxxxx] On Behalf Of 
Walter, Chris
Sent: 03 December 2005 03:41
To: 'thin@xxxxxxxxxxxxx'
Subject: [THIN] Re: OT: NIC Teaming

This is interesting, I have the same NIC's and had similar problems.  I believe 
we have the same switches as well.  We do use teaming but only for failover.  I 
have one NIC set as a standby.  The problem we have with the 100 meg 
connections is that even though the switch is configured for 100/full the 
server side will only connect if I set it to auto even though it shows that it 
connects at 100/full.  I have tried updating the drivers, Bios and others.  
Sometimes resetting the switch port fixes the problem.  I gave up on it though 
because it was happening right before our Q4.  I found that if we switched the 
switch port to 1000/full and set the Broadcom nic to Auto that everything 
connecting ok without errors on the switch port or server and I got my 
performance back.




-----Original Message-----
From: Dirk Blose [mailto:Dirk.Blose@xxxxxxxxxx] 
Sent: Friday, December 02, 2005 5:38 PM
To: thin@xxxxxxxxxxxxx
Subject: [THIN] Re: OT: NIC Teaming



I don't know if anyone has chimed in on this yet but, the most important thing 
with 100MB connections is to make sure the negotiate correctly. The 
recommendation for Server connections is to set them on both the NIC and switch 
port to be what you want such as 100/Full Duplex. If either side is set to 
automatic a negotiation mis-match can occur which can result in a serious 
performance issue. Try that and see if it helps.





Dirk Blose, MCSE, CCA
Lead Technical Analyst
(919) 765-4791
>>> emann@xxxxxxxxxxxxxxxxxxxxx 12/01/05 7:14 AM >>>
Hopefully someone who really knows their stuff about NIC Teams and the
switch side configurations can help me out!  Outside of my particular
issue, I was hoping we could spurr up a convo on what kind of NIC team
everyone uses, against what kind of switches, and for what reasons.

I use all dell servers which have dual Broadcam or Intel GigE NICs.  I
have always teamed them in Adaptive Load Balancing (ALB) mode.  The
requirements for this are basic, and based on my readings, it gives me a
quality mix of load balancing and failover.

Last night I discovered one of my servers had 1 NIC running at 1000 and
the other at 100.  ALB is supposed to work just fine with this.  I
discovered this because I was getting *HORRENDOUS* network performance
from this server.  It was copying 5k chunks of data every 3 seconds or
so, that's how bad it was.

I disabled the NIC connected at 100 and things improved, but only
slightly.  This server is connected to a GigE blade on a Cisco Cat6513.
I checked the port settings and QoS and all look normal.  I'm trying to
figure out why the performance is so aweful.  I'm not sure if it's
related to the team and 1 NIC being down, or something bigger is going

The only requirements for ALB I've read are that no teaming is done on
the switch, and spanning tree if off.  No teaming is done, and I have
spanning tree portfast off for this server, but spanning tree status is
still "forwarding".  I couldn't get any good info on the switch end of
things to know for sure.

Anyone have any tips for me to look into?  I need to do the typical
update NIC drivers, but this is my main production SQL server, so I
can't do it until this weekend, and I know this is really hurting
overall performance out to my users.  I don't have this issue with any
other servers connected to the same switch using the same port setups
that I've seen.

And outside of t
For Archives, RSS, to Unsubscribe, Subscribe or
set Digest or Vacation mode use the below link:

This e-mail has been scanned for all viruses by Star. The
service is powered by MessageLabs. For more information on a proactive
anti-virus service working around the clock, around the globe, visit:

The contents of this transmission are confidential. If you are not the
named addressee or if it has been addressed to you in error, please
notify the sender immediately and then delete this message. 
Any unauthorised copying and transmission is forbidden. Electronic
transmissions cannot be guaranteed to be secure. If verification is
required, please contact the sender.

Other related posts: