High CPU load and strange network problems on IRF-stack

[ulpo]

Hello!

I have to admit I've been a C*sco-user for ages, but now responsible for a medium-sized network built with HPE/Aruba switches.

We have a rather large Vlan 1 with around 500 network-devices on HPE 5130-48G-PoE+-4SFP+ (and similar) with HPE Comware Software, Version 7.1.045, Release 3111P07 configured as an IRF-stack.

<SW-5-1>display arp vlan 1 count
 Total number of entries on interface vlan 1: 438

What I see and experience is

• the network is "unstable" (cannot proof yet, but in daily business we loose pings in LAN)
• I have the feeling that we're experiencing broadcast storms from time to time
• We have (as I think) a high CPU load on all the switches (> 50-60%)

What can I do to track this down in a HP-environment?

Debug-commands, syslogging,...

[ulpo]

Oh. I documented the OS-Version of one of the access-switches.

The core-switch is HP 5800-24G with Comware Software, Version 5.20.105, Release 1809P10 and has "only" 30% CPU-usage.

Thanks.

[Vince-Whirlwind]

1/ You need a topology diagram, "sh cdp ne" is something like "dis lldp inf rem" on 3COM. Map the whole thing out

2/ you need a diagram of your STP topology, including priorities for each switch

3/ you need a monitoring tool to see what kind of traffic you have. In the absence of that, you need to log onto each switch and gain an understanding of how busy its uplink(s) is at min & peaking, as well as which are its busy non-uplink ports

4/ Personally, my first priority would be to eliminate any loops or other immediate performance issues that I would find, then I would very quickly move to segment the network by wiring closet. I have migrated hosts by wiring closet to a new subnet even in the middle of the day - the user will barely notice anything, maybe a brief message from Outlook telling them it's disconnected.
With a segmented network, your performance issues will be much more localised with fewer affected users and much easier for you to pinpoint the cause