Want to protect your cyber security and still get fast solutions? Ask a secure question today.Go Premium

x
  • Status: Solved
  • Priority: Medium
  • Security: Public
  • Views: 315
  • Last Modified:

Cluster Network Activity

I have a 4 Node Cluster using Hyper V and 2008 server. There are two SAN Storage servers. I opened my taskmanger and put it on the network traffic. I logged into each node. I copied a 7GB VM onto the desktop. I did this from each node. All 4 nodes have 2 two iSCSI connections via gigabit network connecting to the san using MPIO. San Has 4 iscsi ports.

On each node I could see the VM copied at about 110mbs and averaged 47% activity on each iSCSI connection.

1. Is that normal? When i only use one Nic it averages 97% and the same 110mbs. Wouldn't two nics be faster.
2.
Why am  not getting 200 mbs?

NOW THE WEIRDNESS
I reversed the test. This time I copied the same VM from each desktop back to the SAN from which it came.  I started with Node 1. I had only HEARTBEAT activity and no iSCSI activity on Node 1. When I looked at Node 4, it had a ton of HEARTBEAT activity and the two iscsi ports lit up and they seem to be doing the work.

1. Why is this happening? I thought the heartbeat was for live migration only. Why didn't node 1 iSCSI ports copy the VM from its own desktop straight to the san. Why did it take the VM from its desktop send it to node 4 through the heartbeat nic to where node 4 used it's iSCSI port to copy it back to the SAN?

Also i noticed heartbeats were at 92% activity but iscsi again was split 50/50.


  This time my HEARTBEAT Nic had all the activity and there was no iscsi traffic.
0
MEATBALLHERO
Asked:
MEATBALLHERO
1 Solution
 
BeartlaoiCommented:
1 and 2. Yes as you have not indicated that you have a real load balancing network setup.  A single TCP converstaion will go through the same ports because there are rules in place, like using the first port that can take the packet where it is supposed to go.
Specifically if you open a command prompt and enter ROUTE PRINT you will see the rules that Windows uses to determine where to send a packet.  
If you want REAL load balancing then you will have to do some serious research into the capabilities of your switches and what client side teaming software to use with it.  LACP for instance will also not accomplish what you want since it explicitely uses a hash based on IP/port on both sides to determine which port to use, so a single TCP connection will thus be stuck on one link.  LACP is designed for trunks that carry multiple conversations, not to load balance one conversation.
Newer switches like Nexus I believe solve this for you.

Other 1.  Do a ROUTE PRINT on both sides and see if you can see why packets may be going in the wrong direction.  One guess, make sure you dont have gateways defined on your heartbeat NICs.  Also if you have seperate iSCSI NICs in addition to the management and VM NICs then they should also not have gateways, only one NIC should have a gateway and that should be management.
0

Featured Post

What does it mean to be "Always On"?

Is your cloud always on? With an Always On cloud you won't have to worry about downtime for maintenance or software application code updates, ensuring that your bottom line isn't affected.

Tackle projects and never again get stuck behind a technical roadblock.
Join Now