NZSolly
Contributor
Contributor

vswitch\pswitch and NIC load balancing

Jump to solution

Hi all, i would like to pass a scenario by you all to check if I am on the right path..

We have 2 hosts with DAS each and the vswitches connect with 2

physical 1GB Nic's to a pswitch (backbone), from the backbone I have 3,

2nd tier switches @ 100/mb which have teamed 1GB connection to the

backbone. With me so far?

Now I believe the network is experiencing request contension in that

sometime we have lag eg: switching shared printers from the print

server, drilling down the network shares etc....

I have check the performance data on the host and i see no heavy

bandwidth, cpu load, disk activity. I have check the NIC's and it

appears only 1 ever gets used even though I am sure they are teamed

correctly.

I have decieded to get another backbone switch same model, and plan

to use spanning tree to connect the backbones and with the NICs

separate them one each to the backbones, and then another spanning tree

to each desktop switch, the idea being that the switching will better

spread the load accross both NIC to the vswitch...

At present there is only the 2 nic's in each host, I have another 2

gb nics and 2 100mb nic to use for the console. I plan to use the 2 new

Gb nic for interhost connectivity and create a virtual SAN from extra

space on each DAS.

I have attached quick diagrams of scurrent and proposed switching and also how I have teaming set up on the vswitch....

Thank you for your time!

0 Kudos
1 Solution

Accepted Solutions
weinstein5
Immortal
Immortal

That is what I get for answering at night - if you change to IP Hash the vmkernel will examine each outbound IP packet and based on the originating and destination IP address the vmkernel wil choose the physical NIC connected to the virtual switch - so for an example and I will keep it simple assume you only have a single VM with a single virtual NIC connected to your virtual switch - now lets look at two possible examples

  1. The VM is a web sevrer provideing conebt to multiple users- each the V's IP address will remain constant but each user will have a different IP address so the outbound traffic will be distributed across all physoical NICs connected to the virtual switch -

  2. The VM is an application server connecting to a physical SQL server - in this case the originating and destination IP addresses will not change so the traffic will only go out one of the physical NICs

Hope this cleared it up -

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful

View solution in original post

0 Kudos
4 Replies
weinstein5
Immortal
Immortal

I think where the confusion is understanding the different types of load balancing methods that ESX provides - ESX provides three types of load balancing - the default which you have selected Rout based upon virtual port ID teh vmkernel choses the physical NIC based upon the virtual port the virtual NIC is ocnnected to - so if you have only a single VM the traffic will only go out a single physical NIC - the second type is the MAC Address based where the vmkernel chooses the physical based on the virtual MAC address and finaly IP Hash where the vmkernel examines all traffic going thorugh the virtual switch and based on the originating and destination IP address chooses which physical NIC to send the traffic out. IP Hash will give you the best probability of spreading the traffic across all the NICs but two things to keep in mind:

  1. The physical switch the Physical NICs are connected to must be able to support link aggregation - Ether Channel in Cisco speak

  2. If the VM is only communicating toa single IP then you will once again only use a single Physical NIC -

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful
NZSolly
Contributor
Contributor

HI, not quite following you, all the VM's only have 1 nic to the vswitch, the vswitch has 2 physical nics to the same physical switch. Are you saying if I only change the teaming to IP Hash, the vswitch will still only use 1 physical nic?....Man, its no wonder I struggle with this stuff....

0 Kudos
weinstein5
Immortal
Immortal

That is what I get for answering at night - if you change to IP Hash the vmkernel will examine each outbound IP packet and based on the originating and destination IP address the vmkernel wil choose the physical NIC connected to the virtual switch - so for an example and I will keep it simple assume you only have a single VM with a single virtual NIC connected to your virtual switch - now lets look at two possible examples

  1. The VM is a web sevrer provideing conebt to multiple users- each the V's IP address will remain constant but each user will have a different IP address so the outbound traffic will be distributed across all physoical NICs connected to the virtual switch -

  2. The VM is an application server connecting to a physical SQL server - in this case the originating and destination IP addresses will not change so the traffic will only go out one of the physical NICs

Hope this cleared it up -

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful

View solution in original post

0 Kudos
NZSolly
Contributor
Contributor

Hi, thanks weinsteins.....that sounds like what I need....I intend to use spanning tree for all the switches so every client on the network will have 2 pathways to the vswitch. I certainly hope this will spread the load better, I have been perplexed as to the lag issue...

Cheers

0 Kudos