Wired Intelligent Edge

I have several 8, 24, and 48 port 2530 series aruba switches experiencing this issue. Upgrading Firmware above YA16.02.0012 or YB16.02.0012 (on the older 8 ports) causes CPU spikes between 25%-30% utilization fairly regularly at about 5 minute intervals. This occurs even when only a single port is connected for monitoring with Airwave, and neither SSH or the web interface is being used during the test period (it's just sitting on my desk. I currently have YA.16.09.0003 (ROM: YA.15.20) installed on a 2530-48G-PoEP (J9772A) and am observing this issue. If I downgrade to YA.16.02.0012, and change nothing else, the CPU utilization drops to a stable rate under 3%. (see Airwave graph screen capture). Seems to be any firmware with the nextgen interface.

Thoughts?

~ tested with YA.16.03.0007, YA.16.04.0008, YA.16.08.0001 - same result
~ tested with minimal setup (set date/time, assign IP address to mgt VLAN) - same result

~ J9772A# display cpu-usage
Master CPU usage:
23% in last 5 seconds
8% in last 1 minute
6% in last 5 minutes

~ J9772A# show cpu

25 percent busy, from 1 sec ago
1 sec ave: 25 percent busy
5 sec ave: 14 percent busy
1 min ave: 7 percent busy

Task usage for last 0 seconds
% CPU | Description
-------+--------------------------
83.2 | Idle
3.4 | Sessions & I/O
10.8 | Hardware Mgmt
1.9 | System Services
0.4 | HTTP
0.0 | IPsec
0.2 | TFTP

Do you have open a issue on TAC ?

no case open yet... figured I'd try the forum first.

it's not a super urgent issue at this point, switches are running fine on 16.02.0012, and the fancy new GUI doesn't really matter to me, but at some point it would be nice to upgrade to current FW for security reasons if nothing else without taking a performance hit. I found a few articles about performance issues (long ping, etc) but nothing about this issue... I posted in hopes that it was just something I overlooked, or that folks might check their CPU logs for similar behavior so I can rule out anything simple. 

Well...there is a previous thread very similar (here) and consider that, originally, Aruba 2530 slowness was discussed here...so, at this point, there could be for sure a relationship between CLI slowness, CPU and Software Versions...great you found that pre-16.03 versions don't show this strange behaviour.

From what I see I can confirm that group of CPU load spikes look like bushes during the day (I worked mostly on 16.08 and 16.09 since ours Aruba 2530 Switches weren't deployed with older ArubaOS-Switch versions...that's to say I never saw the Aruba 2530 going well like Aruba 2540):

Aruba 2530 (today):

As example, again the CPU load through the usual "Today" Performance reports of HPE IMC:

It looks full of bushes...

A comparison with an Aruba 2540 performed few weeks ago - in June 2019 with YA.16.08 - gave these "CPU noise figures":

Aruba 2530:

Aruba 2540:

The Aruba 2540 (even if more loaded - I'm pretty sure of that - in terms of processed data traffic) shows CPU that is only seldom pushed to 30% with veru brief spikes (not bushes of spikes like the Aruba 2530).

Edit: worth to note that it looks like about every 6 hours a group of CPU spikes shows up...

I had looked at that thread earlier on... SSH seems to be fairly snappy, and not really an issue here. My major concern is what it would do to overall performance with all of our switches doing 5 minute cycles of 30%-40% load on top of what actual loading would be... I had also done some latency testing many months ago when I first observed this issue that showed 2 known workstations on the same switch running post 16.03 would nearly double the round trip ping time during these spikes. I've been looking for the screen caps of that testing, but I may have to redo it....

Looking at it from another angle, based on this older article regarding a possible L2 Loop causing high CPU about every 3 minutes, and this one regarding SNMP monitoring causinghigh CPU (I monitor with Airwave), I decided to test the switch completely off network using the system monitor in the nextgen web interface, which provides about 2 hours of visible history. The switch was on the network with a single port all night, I logged into the web interface to check status, and then immediately unplugged the one connected port. I let it run for an hour with nothing plugged in but power, and then reconnected to the network, logged in, and took this screen capture of the system monitor history graph: (bigger version attached below)

It turns out that even with no network connection whatsoever, the switch is still showing 30% peaks about every 5 minutes. The background use is gone, and it's dead flat inbetween, but the peaks are there like clockwork.

I also recall a couple of articles regarding some versions screaming out to aruba central at regular intervals, but according to the switch, all of the settings I can find regarding that are set to disabled. so, another piece of the puzzle....