6

Looking for some advice on our current situation. We have a full rack at a data center that contains 1 uplink to the data center distribution layer switch (we don't control this), 2 HP ProCurve 2824's and a handful of servers (I'll leave it at that to keep things simple). Switch A and Switch B are not fault tolerant. If one fails, we'll lose half of our links. We're trying to configure the two HP 2824's so that if we lose one, normal operation can resume. We currently have trunk ports configured but this is strictly for increasing bandwidth. For example:

interface 17 <br>
name "SERVER-A-BOND0-1"<br>
no lacp<br>
exit<br>
interface 18<br>
name "SERVER-A-BOND0-2"<br>
no lacp

trunk 17-18 Trk4 Trunk

spanning-tree Trk4 priority 4

From what I understand, the switches have to be stacked (Switch A is the commander and Switch B would be a member - this is ProCurve lingo by the way), and what we're trying to do is called InterSwitch Trunking, although the ProCurve manual doesn't refer to it as that.

I'm assuming once the switches are stacked I can do a similar configuration except interface 17 above would be interface 1 on Switch A and interface 18 would be interface 1 on Switch B. If Switch A or B fails, there won't be a single point of failure. The ProCurve manual specifically mentions using LACP for redundant switches and the "trunk" option for when interoperability is an issue. What I have above me need to be switched to something more like

trunk 17-18 Trk4 LACP

Would really appreciate any comments as to whether we're going down the right path with the new configuration. Unfortunately, the only environment we have to work with at the moment is the production environment which makes testing different scenarios difficult.

Lastly, for the switch uplink we currently have 1 drop from the data center. For this to work we'll need two and I'm assuming that the distribution layer switch (where these drops come from) will need to be configured for link aggregation/LACP, so there will need to be some changes by the data center staff.

Chida
  • 2,471
  • 1
  • 16
  • 29
aj4525
  • 61
  • 1
  • 2

2 Answers2

3

You can split a bond across the switches, but not to increase throughput. So if a server has two NICs and you plug one into each Switch, then you'll have to configure the server for some sort of bonding with failover only.

This Q/A has a lot of good details: Server-to-Switch Trunking in Procurve switch, what does this mean?

The switches have to be plugged into each other, obviously; but the single uplink to the datacenter is going to be a problem. You'll really need to get a second uplink or you'll always have a 50/50 chance of losing your connection when a switch dies (or stops working for any other reason).

Chris S
  • 77,337
  • 11
  • 120
  • 212
  • 1
    I had the same problem, here you have solution but we have a 2 uplinks to DC http://blog.adamczyk.czest.pl/2011/08/switch-redundancy-with-bonding-2/ – klocek Aug 17 '12 at 14:18
  • Thanks for the response. Since my servers already have bonded interfaces, I should just be able to get a second drop from the data center and then connect Server A NIC1 into Switch A and Server A NIC2 into Switch B. I've been getting some mixed responses on this issue from other resources, I feel like I'm just going to ultimately have to test it out. – aj4525 Aug 17 '12 at 14:26
  • What OS and NICs are you using? I may be able to get more specific instructions. This setup is almost exactly what we use in our racks. Works well for us, but if you need >1Gb you may get into some interesting configurations. – Chris S Aug 17 '12 at 14:37
  • I read both links and think I understand how this would work. We're actually in the process of migrating from XenServer to VMware ESXI 5.0. The NICS are Broadcom 5709s. It sounds like IST for HP is called "Distributed Trunking" and this is where you could connect NIC1 into Switch A and NIC2 in Switch B and you would have a combination of more band width and fault tolerance. Reading the links above, it sounds like you can do this on the server side having one nic being active and one being backup. This would provide fault tolerance but not allow for more bandwidth. – aj4525 Aug 17 '12 at 14:43
  • Generally correct. If you need the additional bandwidth, and want redundancy, you'll have to change out the switches for something more advanced. – Chris S Aug 17 '12 at 14:54
-1

It looks like splitting a trunk between two 2824's is not possible. This functionality starts in the 3800 series models (and probably above). The 2824 can stack but only for administrative purposes (not "true" stacking).

aj4525
  • 61
  • 1
  • 2