1

We are experiencing slowness and I am fairly new to using SANs so I would like some help with this. First off, I don't think this will fully solve our issue, but I would like to focus on our iSCSI connection for now. We have 3 PowerEdge R710s for servers and 2 PowerVault MD3220is. They are connected by 2 PowerConnect Gb (one on 130.x and one on 131.x) switches and each has a spare NIC.

All 3 ESXi hosts pretty much all have the same setup:

vSwitch1 (bound to NIC1)
vmk1 IP: 192.168.130.1

vSwitch2 (bound to NIC2)
vmk2 IP: 192.168.131.1

Each PowerVault has 2 controllers with 4 NICs for 8 NICs total. They are pretty much identically configured as follows:

Controller 0/1: 192.168.130.101
Controller 0/2: 192.168.131.101
Controller 0/3: 192.168.132.101 Unused
Controller 0/4: 192.168.133.101 Unused

Controller 1/1: 192.168.130.102
Controller 1/1: 192.168.131.102
Controller 1/1: 192.168.132.102 Unused
Controller 1/1: 192.168.133.102 Unused

Is this the ideal configuration? It seems like we could get more throughput if we put everything on the same network like so:

vSwitch1 (bound to NIC1)
vmk1 IP: 192.168.130.1
vmk2 IP: 192.168.130.2
vmk3 IP: 192.168.130.3 (adding the unused NIC)

For the PowerVaults:
Controller 0/1: 192.168.130.101
Controller 0/2: 192.168.130.102
Controller 0/3: 192.168.130.103 (adding unused NIC)
Controller 0/4: 192.168.130.104 (adding unused NIC)

Controller 1/1: 192.168.130.105
Controller 1/1: 192.168.130.106
Controller 1/1: 192.168.130.107 (adding unused NIC)
Controller 1/1: 192.168.130.108 (adding unused NIC)

I would wanted to put every other port on switch1 and the rest on switch2.

Would that provide enough redundancy? Would it speed things up? Is there a better way?

Philosophene
  • 202
  • 1
  • 12

2 Answers2

1

I'm having a related problem where I can't use the full bandwith, however I can help you with your question. To balance the iSCSI traffic across both switches you should use different subnets. Please see my posted image here: iSCSI with 3 hosts

Using one subnet for all NICs would make it hard for the iSCSI initiator to determine which target to use, I guess. In your case I would suggest to change the configuration to:

Controller 0/1: 192.168.130.101
Controller 0/2: 192.168.131.101
Controller 0/3: 192.168.130.111
Controller 0/4: 192.168.131.111

Controller 1/1: 192.168.130.102
Controller 1/1: 192.168.131.102
Controller 1/1: 192.168.130.112
Controller 1/1: 192.168.131.112

Community
  • 1
Ryan Hardy
  • 210
  • 2
  • 6
0

We are experiencing slowness

You need to quantify your slowness. High latency? Low throughput? Different problems have different solutions.

It seems like we could get more throughput if we put everything on the same network like so

Is there any basis for this? Or is it just a gut feeling coming from last night's curry?

You should be treating your iSCSI SAN the same way that you'd treat an FC SAN. Two independent networks on independent switches. What you have now is the right approach.

With your current setup each host should see 4 paths to the storage:

  • vmk1 → c0/1
  • vmk1 → c1/1
  • vmk2 → c0/2
  • vmk2 → c1/2

You can improve bandwidth available out of the iSCSI storage array by hooking up ports 3 and 4 to the "first and second" networks (as Ryan Hardy suggests) and adjusting which ports are exposed to which hosts (balance them).

MikeyB
  • 38,725
  • 10
  • 102
  • 186
  • The suggestion for putting everything was actually from a VMware tech, although I do not think it was a proper suggestion. Slowness is in the form of high latency. – Philosophene Jun 19 '14 at 16:06
  • High queueing latency? High network latency? High transaction processing latency on the subsystem? – MikeyB Jun 19 '14 at 17:27
  • According to ESXi: command latency, phsical device command latency, physical device write latency, write latency, read latency, physical device read latency. – Philosophene Jun 19 '14 at 21:07