Unsolved
This post is more than 5 years old
5 Posts
0
24111
Help needed with NIC Teaming on Windows Storage Server 2008R2 through PowerConnect 2724
We have a Dell Powervault NX with 2 Broadcom NICs (4 ports) as our primary file server connected to our LAN through a PowerConnect 2724. I wanted to increase throughput to it so I began looking into teaming the NICs with Broadcom's Advanced Control Suite.
I have created an 802.3ad LAG using LACP team in the software wizard with all 4 ports. I manually assigned the LAN info on the new team NIC and the file server is up and accessible...but not without issues. I am getting some intermittent delays when accessing the shares on that server (or hiccups during RDP sessions to it).
When you select that team type, you get a warning to make sure the switch is configured correctly. Not being a "network guy" I assumed that meant creating a corresponding LAG in the switch for the 4 ports the server is connected to. When I did that, communication with the static IP for the server stopped. When I remove the ports from the LAG, I am again able to communicate with the server.
So, question #1 - is there anything special I need to do in the 2724 to support the NIC Team on that server like create a LAG with some specific setup or change how the ports are configured individually? Right now, those 4 ports are set to:
- auto negotiate- disabled (I read that was necessary somewhere)
- full duplex 1000 Full
- backpressure - disabled
- flow control - enabled
Question #2 - in the Broadcom software/team view, i see the 4 adapters listed in the team. One of them has a different ICON (the team icon) and the other three are different. Not sure if this matters or not, but related to that, when I look at the TX/RX stats, the NIC with the "team" icon is over 300million frames TX and the other NICS are under 1000 frames. To me, that seems indicative of something being unbalanced.
Question #3 - is the team type I used my best option (assuming I can get it working correctly) for throughput or would I be better suited using either the Smart LoadBalancing or Generic Trunking team types?
Question #4 - I have all 4 ports plugged into the one switch (I'm not looking for redundancy as much as throughout, so I read this was the better option). I have a second 2724 and a 2848 available as well. Should I stick with all 4 into one switch or vary this? Again, throughput being the primary goal here.
Question #5 - i have a PowerConnect 2848 I can use instead if for some reason that switch is better suited to handle this.
Thanks for any help on this! Much appreciated
bcontento
5 Posts
0
March 18th, 2015 10:00
I may have found a partial answer myself. I just read that the 27xx and 28xx series PowerConnects do not support LACP. They only support 802.3ad STATIC . That would certainly explain part of the problem I would think.
I see in the Broadcom Team setup screen that I can chose Generic Trunking/802.3ad-Draft Static . Is this what I need to do? And then create the 4-port LAG on the switch?
Anonymous
5 Practitioner
5 Practitioner
•
274.2K Posts
0
March 19th, 2015 06:00
Your findings are correct. The LAG on the 2724 is a static LAG. So you will need to pick the static option when creating the Broadcom team. Most of the time there is also a switch independent teaming option. You might try both out and see which one works best for your environment.
http://bit.ly/1ocsJB6
bcontento
5 Posts
0
March 19th, 2015 07:00
Thanks for the reply.
I did some work on this last night:
- I created 2 LAGs on the 2848, 4 ports each. each LAG connects to one of the 2724s where I created the matching 4 port LAG. I turned off auto negotiation on all of the ports and LAGs
2724=====2848=====2724
===== ======
- I created a third 4-port LAG on the 2848 for the Generic Trunking LAG from the file server (converted the LAG from the 802.3ac LACP to generic trunking static). Now, the stats on the 4 adapters are MUCH more evenly balanced (and they all have the same 'team' icon)
- I patched as many of the more critical machines in the network (servers, wifi routers, heavy use desktops, etc) into the remaining ports on the 2848 since it's now my central switch (all servers and it hosts both trunks). I figured this would reduce the connections that a good portion of the network traffic needed to traverse
how does all of that look/sound? anything in particular that I should try, remedy, change, set on the switch(es).
I found a little program called LAN Speed Test (lite -free version). It allows you to enter a network share, chose a packet size and it will write a file and read a file of that size to and from the share. When I run that from my laptop (gigabit wired no wifi) to the main file server (with the Broadcom team), I get uploads speeds of a little more than 600Mbps and download speeds of around 675Mbps. From one of our high-performance workstations I get like 780down and 650up. I was hoping for more...am I being unrealistic?
Anonymous
5 Practitioner
5 Practitioner
•
274.2K Posts
0
March 19th, 2015 12:00
The setup sounds fine to me. With link aggregation you are not always going to see single connection speed increases. Kind of like a 4 lane highway is going to let more cars do the speed limit Vs. a single lane road. This document is a little old, but still has some good info.
www.ieee802.org/.../frazier_01_0407.pdf