4 x 2920-24G | two racks | redundancy stacking or not

[linuxmail]

hi,

I'm new to deeper networking related stuff and HP switches too :-), so I have a couple of questions:

We have in our datacenter two racks for our hypervisors (Xen/Proxmox/...) and normal hosts. Actually we have for the productive servers two LANCOM switchs (german manufacture) in every rack (two for internal communication, two for external communication) and they are connected via lacp and two ports. So if one switch fails, all servers could reach the targets via the remaining switch.

Now we want to replace the 4 LANCOMs with 4 2920-24G  (J9726A) in stacked mode. My first thought was, to connect two switches in the same rack via a 0.5m stack cable, and for the rack connections the longest stack cable (3.0m), to get a ring topology ( so, 2 x 0.5m and 2 x 3.0m).
But two points wherer unclear:

• Do I get in trouble with latency, because of two types of stack cable length?
• What, if I have to update the firmware. I red, that the whole stack must be rebooted and than, all servers are offline.

That is bad :-)

So, what is the preferred solution to solve that problem? Maybe I stacking only the two switches, which are in the same rack and create a LACP/TRUNK connection with the other switches in the other rack?

cu denny

[Vince-Whirlwind]

The different cable lengths to create the IRF should be fine. At one point, they were selling 5m, 7m, and even 10m cables, but I'm pretty sure they stopped selling the longer lengths, and that may have been something to do with the problem you are hoping to avoid.

I had a quick glance at the ISSU information in the 2920 manual and it doesn't say anything about rebooting switches individually. As you say, this is bad.

For a server setup, unless you are allowed to have occasional outage windows, then using a single stack may not be the best way to go.

I *think* the 2920 supports distributed trunking, though, which may offer you an alternative setup for trunk redundancy.