Hi Lee, No I don't mind opening this up a little. I believe Aruba Networks have a very good understanding of the limitations imposed during an ambitious scaling up of multiple wireless networks.
We run Aruba network SUP2 controllers in model 6000 chassis. We have also moved a few controllers to the M3 controller architecture. SUP2 has 3 CPUs and 256M of flash RAM (max. 250 APs, now derated to max. 170 APs) M3 has 1 powerful CPU (spec.s? and a separate crypto chip) and 1Gig of memory (was max. 512 APs, they may come out with a new spec) So, the M3's hardly caused us to break a sweat. The SUP2's under load and with 8 SSIDs (4 on each band) would start to randomly reboot. This was due to the watchdog process catching a low free memory condition and reloading the controller. I soon put together some SNMP monitoring scripts and used our MRTG graphs (CPU and Free Mem.) to predict an impending reload and scheduled a maintenance window to reload it ourselves at an appropriate time. Finally, I would see our CPU1 of the SUP2 controllers spike to between 80-100%, although I am not aware of any client wireless experience degradation. I don't remember what is controlled by CPU1 (data, control plane, ...). I would say that Aruba worked hard at finding issues and memory leaks in the code. They managed to make enough optimizations for us to continue through our school year without having to make major hardware changes, while maintaining 7 SSIDs across 2 bands with a minor down scaling of Max APs. I had to scale down some ARM features as well (band steering was disabled). Of course, with the M3 controllers in place we would be able to run all the ARM features. The M3s on our campus network have the 7 SSIDs, one of them covers all the Libraries and has regularly seen over 2000 simultaneous client connections with the CPU rarely reaching 10% and free memory never dropping below 90% free. I wonder if I could double the SSID count without the M3 complaining. Not that we need 14-16 unique SSIDs. We will upgrade to all M3s on campus this summer. I will let you know if I manage to break it ;) Code 3.3.2.24 (end of code branch) You asked what memory I am referencing. I am not sure how to answer that. I am only aware of one value to monitor. Unless you mean CPU caches? The Aruba box is Linux with the Aruba code, 256MB flash for SUP2 and 1 Gig for M3. Maybe someone on the list could help me hear. Let me know if missed something, and feel free to contact me offline. Cheers Lee, John Wheeler Wireless Network Specialist Network Infrastructure Team - McGill NCS -----Original Message----- From: The EDUCAUSE Wireless Issues Constituent Group Listserv [mailto:wireless-...@listserv.educause.edu] On Behalf Of Lee H Badman Sent: Thursday, May 06, 2010 5:23 PM To: WIRELESS-LAN@LISTSERV.EDUCAUSE.EDU Subject: Re: [WIRELESS-LAN] Dual SSIDs John- Which memory and CPU are you referencing? And do you mind saying which manufacturer and models/code you are using? -Lee Badman ________________________________________ From: The EDUCAUSE Wireless Issues Constituent Group Listserv [wireless-...@listserv.educause.edu] On Behalf Of John Wheeler, Mr. [john.whee...@mcgill.ca] Sent: Thursday, May 06, 2010 5:06 PM To: WIRELESS-LAN@LISTSERV.EDUCAUSE.EDU Subject: Re: [WIRELESS-LAN] Dual SSIDs Hello, My 2 yen: I don't know if this was covered, sorry if it is repetition. Some architectures have GRE tunnels for each AP, band (2.4 & 5 GHz), and network (number of SSIDs). This overhead tends to build up fast. If you have 2000+ dual band APs with 2 SSIDs, you will be handling 8000 GRE tunnels. Add an SSID or two across the board and watch your free memory and CPU start to creep into uncomfortable areas. Having said that, we run 4 SSIDs on our network (wpa/wpa2, captive portal, Eduroam, and Guest). One of the networks had to be removed from 5GHz to accommodate the load on that CPU. New Controller hardware will help give us some more breathing room. I would say that the GRE tunnel limitation has affected us more than SSID overhead of answering probes and beacon transmissions. Although, I may not be looking at the SSID overhead problem closely enough. Our campus network is running with over 2200 APs in the downtown core of the city. On any given day I can see 6000-11000 foreign SSIDs. It was only when we double the SSID count under a beginning of semester load that our controllers slammed into the wall. Guess who is no longer on 5GHz ;) Cheers, John Wheeler Wireless Network Specialist Network Infrastructure Team - McGill NCS (514) 398-7388 john.whee...@mcgill.ca From: The EDUCAUSE Wireless Issues Constituent Group Listserv [mailto:wireless-...@listserv.educause.edu] On Behalf Of Greg Williams Sent: Thursday, May 06, 2010 2:40 PM To: WIRELESS-LAN@LISTSERV.EDUCAUSE.EDU Subject: Re: [WIRELESS-LAN] Dual SSIDs Hello, My two cents: As people have already pointed out, the limiting factors are the overhead from beacons as well as responses to client probe requests. When you consider that each BSSID will be transmitting beacon frames as well as responding to client probe requests (and each AP will respond simultaneously to these probes typically), and that in a typical deployment you have multiple overlapping AP coverage areas (especially in 2.4GHz with greater penetration, which also has only three non-overlapping channels).... you get the point. The overhead from this traffic can add up very quickly. That being said, depending on 1.) how many APs are overlapping in a given area and 2.) whether or not all of the APs are set on the same channel or not in a given area and 3.) the number of clients in the area, we have found that 4-6 SSIDs is where noticeable performance drop-off begins occurring. Hope that helps. Greg Williams Product Manager 415.203.5630 meraki.com<http://meraki.com/> ********** Participation and subscription information for this EDUCAUSE Constituent Group discussion list can be found at http://www.educause.edu/groups/. ********** Participation and subscription information for this EDUCAUSE Constituent Group discussion list can be found at http://www.educause.edu/groups/. ********** Participation and subscription information for this EDUCAUSE Constituent Group discussion list can be found at http://www.educause.edu/groups/. ********** Participation and subscription information for this EDUCAUSE Constituent Group discussion list can be found at http://www.educause.edu/groups/.
