Problem with Network Teaming and Receive Load Balancing

[Thomas VIE]

I've an ML370G4 with high incomming traffic. There are 2 Network-Adapters (GBit) inside and creating the Team was no problem (Switch assisted load balancing) - the Server is connected to a Gbit Cisco Switch (3560) and the ports are in trunk-mode. I can see in the HP Network Software both adapters with 1 GBit RX and TX and in the utilization i can see that he supports up to 2 GBit RX and TX.

The problem is now, that i don't get more then 1 Gbit incomming traffic - it's just 1 adapter then used at nearly 100% and the other one is not realy used. Sometimes Adapter 1 is used and sometimes Adapter 2, but never a load-balancing on both adapters or more then 1 GBit incomming.

I already updated the HP Networking Software but this didn't help.

[KarloChacon]

hi

do you have the PEINP license (required Switch assisted load balancing)?

check the document attached, page 4, also you can check the configuration

more info here
Switch-assisted Dual Channel Load Balancing ((Advanced Pack).
Switch-assisted Dual Channel Load Balancing requires activation with a license key and is part of the INP upgrade for HP ProLiant Network Adapter Teaming.
There must be a minimum of two adapters in a dual channel team and a minimum of one adapter per group. One group of team members is treated as a Switch-assisted Load Balancing (SLB) group to one switch, and a second group of team members is treated as an SLB group to a second switch.
All transmit packets are load balanced among all team members based on a Load Balancing
algorithm in the teaming device driver. The receive packets are load balanced among all team members by the switch. If a failure of any team member occurs, the packets are load balanced among the remaining adapters. Only Gigabit adapters can be used for Switch-assisted Dual Channel Load Balancing. When you select Switch-assisted Dual Channel Load Balancing, the Team Members window becomes Team Members Grouping and displays two windows labeled Group 0 and Group 1. The up and down arrows are used to assign the members to be in Group 0 and Group 1. If a team contains an iSCSI-enabled adapter, you cannot use the Switch-assisted Dual Channel Load Balancing (INP) setting.

regards

[KarloChacon]

hi

you say you have the latest NCU, but what about latest drivers and firmware for the NIC Cards?

regard

[Thomas VIE]

Whats the PEINP license?

I used the PDF for Network Teaming and there is just something written about the Network Intelligent Advanced Pack if you like to use SLB across 2 switches - if you stay on one Switch, just the Teaming of the 2 Network Cards and setting the 2 ports on 1 Switch on Trunk Mode should be enough.

I don't need redundancy, thats why connection to two Switches is not requred - i just need more bandwith for incomming traffic.

[KarloChacon]

hi

check here about the license to use advance network capabilities

http://h20000.www2.hp.com/bc/docs/support/SupportManual/c00287087/c00287087.pdf

as you read on the whitepaper
"Switch redundancy is required" and the switch must support that feature

regards

[Thomas VIE]

But thats about Dual Channel SLB where you need 2 Switches, 4 Networkcards, ... - I just have 2 Networkcards and connect them both to the same switch.

Here you can see what i mean - it's the simple SLB and not the Dual Channel - ftp://ftp.compaq.com/pub/products/servers/networking/capability.pdf

[Thomas VIE]

Here is the grafic for what i mean - page 18, the option on the bottom left:
ftp://ftp.compaq.com/pub/products/servers/networking/TeamingWP.pdf

[KarloChacon · Accepted solution]

sorry

I did not read very well, so this is the one:
Switch-assisted Load Balancing with Fault Tolerance (SLB)

OK so you get the 2 GB of Tx? but just (almost 1 GB on Rx - incoming)

just to be sure that switch support SLB? has the latest ISO?

http://www.cisco.com/en/US/products/hw/switches/ps5023/prod_release_note09186a008089a797.html

check TeamingWP.pdf, page 78 question 2
Why is traffic not being load balanced into my server?
A team type of SLB, 802.3ad Dynamic or Dual Channel and a supporting switch are needed to achieve receive load balancing. Receive load balancing is determined by the switch connected to the HP team in SLB mode. If the HP NIC team is configured for SLB, then receive load balancing should occur if the switch is configured properly. Consult the technical resources provided by the switch manufacturer.

[Thomas VIE]

I don't know if i get 2 GBit TX because there is not realy TX on this box, just heavy RX but that with the switch sounds comprehensibly - i'll check the switch if it's realy completly correct configured (possibly just to set trunk is not enough?).

[KarloChacon]

hi

also check TeamingWP.pdf
page 58 to get more feedback about SLB (switch-assisted load balance not server load balance)technologies

regards

[Mi6t0]

Hello, Thomas.

And did you try setting it to Transmit load balancing? Is it the same behaviour?

[hpabc]

Hello,

I have set it to Transmist Load Balancing with Fault Tolerance. But i am also facing same problem. I am not getting RX on one of the two NICs and due to which the users connecting to this server are being disconnected for few seconds by itself and connects again, which causes the user PCs to hang.

I have set the speed/duplex to 1000/full and set the speed on switch to 1000 manually on both ports.

While copying 500MB of files from and to server is similar, like 1 min apprx.

The information shows one of the NIC role as Tx/Rx with Group- 0 / Member id- 0.
The information of the other NIC role as Tx Only with Group- 1 / Member id- 1.

Also the peak usage of the Rx in percent is nearly 1-2% and sending is hardly 1%.

I have not installed the latest drivers. Could this resolve this problem ?

Please guide.

Thanks,

hpabc.

[Thomas VIE]

@mi6t0: in the current configuration i have 2 GBit RX and TX, Transmit Load Balancing would be just 1 GBit for RX and 2 GBit for TX.

@hpabc: with the setting for Transmit Load Balancing one of the Adapters work for TX Only and just one for RX and TX - the adapter with TX only will not get any RX traffic in this configuration.

[hpabc]

Hi,

Thanks for reply.

Yes, with TLB the receiving traffic is not load balanced. The Primary teamed port is responsible for receiving all traffic destined for the server.

So things does not look awkward with TLB settings on the NIC. But users complain slow connections sometimes. I am out of ideas.

Thanks

hpabc.

[hpabc]

Hi,

Thanks for clearing that.

Yes, with TLB the receiving traffic is not load balanced. The Primary teamed port is responsible for receiving all traffic destined for the server.

So things does not look awkward with TLB settings on the NIC. But users complain slow connections sometimes. I am out of ideas.

Thanks

hpabc.

[KarloChacon]

hi

I was doing more research about this capability on Cisco switches which called Fast EtherChannel (FEC) mode or Gigabit EtherChannel (GEC) that let us to use SLB
here is the information with the cisco 3560:

http://www.cisco.com/en/US/docs/switches/lan/catalyst3560/software/release/12.1_19_ea1/configuration/guide/swethchl.html

regards

[Connery]

Hi Thomas,
I'm an engineer from the development group responsible for ProLiant networking products (i.e. NIC Teaming).

Two things to check: 1) switch's channel configuration 2) switch's load balancing algorithm.

1) In order to achieve receive load balancing, you need to verify that your switch is setup for EtherChanneling.

You mentioned that the Cisco is configured for Trunking, however, on the Cisco switches the term "Trunking" refers to "VLAN tagging". "EtherChannel" or just "channeling" refers to bonding multiple ports together for load balancing.

On the ports in question on the Switch, your configuration should look something like this:

(assumes you are using IOS, switch ports are ports 10 & 11, and unused port channel number is selected).

interface range 0/10 - 0/11
switchport mode access
channel-group 10 mode on

2) If you have confirmed that the channeling configuration is correct and you are still not achieving receive load balancing, the problem may be the switch's load balancing algorithm. Make sure the switch is set to load balance by source & destination IP address.

My guess is that most, if not all, of your clients are on the other side of a router from your server. If that is the case, then all of the packets from the clients pass through the router to get to the server. As the packets pass through the router, the router encapsulates them inside of ethernet frames (something we all already know). Anyway, the router always inserts it's own MAC address as the source MAC address of all frames (from all clients). In other words, all client frames come from the router's MAC address. Since the 3560's default load balancing algorithm is to load balance by source MAC address, all frames coming from the router are sent down a single port in the channel. As a result, all client traffic is transmitted to the server over a single link in the channel group. To effectively load balance the client traffic, the switch needs to be changed to load balance based on Source & Destination IP address. That way, as long as either the source or destination IP address changes, a new port in the channel can be selected and load balancing can be achieved.

To change the load balancing algorithm to be based on Source & Destination IP address, use the command "port-channel load-balance src-dst-ip" in Global Configuration.

Feel free to attach your switch config if you'd like us to verify the configuration. Be sure to include the port numbers you are using.

Best regards,
-sean

Network Architect, CCIE #18040

[philippjaeger]

Hi Sean,

are you still available?

I have a customer with a very similar situation/problem regarding nic teaming and not getting any Receive traffic load balanced over both NICs.

Customer test traffic is only between ProLiants connected to one ProCurve Switch. Can you let me know what config data you exactly would need in order to figure out why traffic sticks to one NIC only all the time?

Thanks & Regards,

Phil