10.2r4 firmware causing AP650's to lose connectivity?

[KevinP1]

Okay, I don’t know what’s happening here but this isn’t looking pretty. All of our AP650’s are losing connectivity multiple times throughout the day. 

Look at this! Some AP’s will only do this once or twice and others are doing it often, like this ? There’s nothing indicating any type of error or issue. We didn’t have this problem on firmware 10.09rb.

I have a case open with GTAC, but are we the only ones seeing this issue?

[jose_gonzalez]

Some other tips to improve performance off the top of my head:

• QOS is software based, takes a big hit on CPU, especially on older hardware, even if you haven’t defined any QOS policies, just having the engine enabled reduces performance.  We would not be operational today if we didn’t use ‘no qos enable no-prompt’ in CLI supplicant.  In the old HiveManager disabling the QOS engine was referred to as ‘Turbomode’ in GUI, no joke.
• In the radio profiles, by default 'Enable background scanning when there are clients connected’ is on, disable this.  Clients have to be disconnected in order for background scanning to occur.
• In the radio profiles, if you have switching channels enabled on interference thresholds, don’t ‘Enable channel switching even if stations are connected’.  I have observed APs switching channels every couple of minutes because of this, re-trying the same channels over and over.
• Disable proxy-arp if enabled, caused issue for me in the past, not needed.
• Disable the service console if enabled, I have observed APs go into this mode inappropriately.  It disconnects clients off the AP when in service mode.
• In 1-to-1 and dense areas take advantage of the DFS channels on the 5ghz band.
• Take advantage of features to improve performance, enable aggregate MAC protocol data units.  On the 5ghz enable frame burst, MU-MIMO, short guard interval (not outdoors),
• If you do load balancing between APs, don’t use client based, use airtime.  Client based is buggy, I have observed clients being denied access because of this.
• Disable power management wifi settings on the clients if possible.

[KevinP1]

@dstarkman Yes, 10.3r1 has been working well for a few weeks. It does sound like your laptops are turning off the WiFi adapters to conserve power. If that has been disabled, look into if there’s interference in the area (SSH into the AP and the ‘show station’ command), roaming settings, and power settings on your AP’s. If you’re able to provide me with some info from your device template(s) and your Optional Settings for your SSID; maybe we can get this figured out for you.

[dstarkman]

@kevin.piazza  You got 10.3r1 to work? I was trying it out in my lab where there was only two laptops connected to it and over the weekend they just dropped off. I thought maybe since these laptops had the lids closed maybe they had a power save issue. I opened the lid and tried to reconnect and they wouldn’t. I rebooted the AP and they came back alive. The next day I came in and they were disconnected yet again. Same thing had to reboot the AP for it to reconnect. I just rolled it back to 10.0.10.2 which seems to be fine in my office but it’s got issues in my production area with multiple entries for the same time for connects and disconnects which makes the logs really hard to go through.

[jose_gonzalez]

I read through and saw pages of users worrying of memory usage dips trying to correlate to performance problems.  The web GUI is pretty, but has never been reliable.  I have never observed big spikes in either direction with memory usage.  If you want to graph reliable metrics to CPU, memory, and client count you have to enable SNMP and use 3rd party applications.

OIDs to graph:

CPU: 1.3.6.1.4.1.26928.1.2.3.0
Memory: .1.3.6.1.4.1.26928.1.2.4.0
Client Count: .1.3.6.1.4.1.26928.1.2.9.0
 

Sample AP metrics graphed via SNMP

[SamPirok]

Just something to consider, the auto power and channel adjustments are based on background scans the AP makes; if those aren’t going through then the settings won’t be adjusted. Also, I’m told they can only take in to account signals from other Aerohive devices when they make these adjustments, which isn’t the whole picture when it comes to the airspace your signal is in, so the auto selection settings have some pretty significant limitations. The places I’ve seen power levels of 1 or 2 are typically schools or hospitals where we have an AP in almost every room, so we need to reduce the distance the signal is traveling to reduce overlap. Just my 2 cents on those auto settings.