I am trying to figure out the best way to improve reliability and uptime of my network in my server rack.
Facts:
– Sonicwall 3600 FW/Router (active/passive) – 1 internet and 1 leased line connection, + 2 IP sec tunnels
– 2x HP2920 switches with stacking module (only 1 switch is actually in use right now)
– about 15 host machines (2 NIC) running VM's (probably about 50 VM's or so)
– Network is VLAN'ed (flat segment for the physical machines, various VLAN's for the VM's)
It's a production environment so I can't make major changes, and can't spend any more money, so I am trying to see how to make best use of what I have.
We tried to configure NIC bonding at the server level to connect out 1 NIC to 1 switch each , but that caused a ARP flood that brought down the whole network, so obviously we weren't doing things right. Hosts are Unix hosts. Anyone has experience in this kind of setup to provide some guidance? We are not a big shop to have all the complex networking setups that you find when Googling around – just trying to get something relatively simple that meets our needs so that I don't have hours and hours of outage if the switch goes down.
Best Answer
If you have the HP2920's stacked together and they are being seen as one switch fabric, you can plug an Ethernet cable to each switch from each server and use LACP trunks to setup link aggregation.
I understand that when you did this last time your network went down but this will more than likely be to the fact that you did not setup an LACP trunk on your switch. You will need to make sure you configure this on the switches before connecting them to the servers.
If you had LACP configured on the server but not the switches, that the sort of behaviour I would expect to see.
You can start of with just one link plugged in for continuation of service and add the other once all is configured. Just be aware that there will be a brief outage while the links are aggregated. This should only be a few seconds.
To configure the trunk on a procurve switch, use the following command:
This will tell the switch to send LACP control packets and try to establish a LAG between the switches and the server. If your server does not support LACP, you will need to use something like
Which will establish a static LAG and force one to be created.
Using this method you will receive redundancy between your switch and servers.
I hope this helps you on your quest.
SleepyMan