Wanted to give an update on this issue and reply to a couple of the
responses we've gotten back:
- first of all, I do want to point out that Airwave has been
tremendously helpful in working with us on these issues; they've come up
with a number of database tweaks that have helped some. Most
importantly on Friday afternoon, they identified a specific performance
problem with the Search feature and developed a patch for this issue.
They believe this patch will be in 6.3, but they don't know for sure.
Page loads went from around 40 seconds to 10 with this patch when they
applied it to our server. So, kudos for Airwave support for all the
effort they've put in to working with us on this
- secondly, folks have asked about the 5 minute polling period and/or
using WCS instead. Two issues there: (1) it is not necessarily safe
to assume that a WCS environment would be our preferred wireless
architecture in the long term (2) we are attempting to get a complete
set of all user related statistics, not just authentication messages, as
often as possible. With more and more users requesting the same level
of support with wireless as they do with wired, we want data as close to
real-time as possible. While we can get a lot of this from the
controller, having to collect info from as few applications as possible
is appreciated. What really drove the performance evaluation was a tool
that we developed that allows users to find out all the detailed
information about their session (wired or wireless), which can currently
be frustrating as Airwave may not show wireless users associated for up
to 25 minutes after they connect to the network. And at that point all
the data is pretty old...
- and finally, why we don't split it up into multiple Airwave servers:
we tried that and it was not seamless. Their idea of a "Master
Console" isn't the same as ours and it's not really a clean distributed
architecture where you can hit one console to get the consolidated
information.
Thanks for the feedback to date!
-- Jim Gogan
Director, Networking
ITS Communication Technologies
UNC-Chapel Hill
Jim Gogan wrote:
We have been dealing with some Airwave performance issues for quite a
while, and wanted to compare results with others out there that have
Airwave installed and use it to support their user base. Any
information would be very helpful.
We have about 1800 access points across about 30 controllers. The vast
majority of these APs are thin Cisco with some small installs of Aruba
(less than 100). We peak around 8500 concurrent wireless users during
the fall and spring session.
Our problem has to do with polling times for User Data, and the expected
time for newly connected machines to be reflected as associated within
Airwave.
We want nothing less than 5 minute polling period for user data. We
have set everything else to 30 minutes in order to reduce load on our
server (which is a dual quad core Xeon with 16 gigs of RAM and (2) 15k
RPM SAS drives in RAID 0). We are seeing what we would consider less
than stellar results.
During the school year, it would take this box, at times, over 20
minutes to actually show a user as associated. That means, as we
understand it, the data for troubleshooting the connection is 20 minutes
old. Given our expected level of service and support, this is not
acceptable.
We are expecting that the Airwave server should show a new user within 5
minutes for the majority of the time, and be no more than 7 or 8 minutes
worst case (explained: if the server just finished polling a controller,
and a new user connects, then it would take 5 minutes for Airwave to
discover this device, and another 2 to 3 minutes to parse the data. If
the user connects just prior to a poll, then it would be possible to see
the user prior to 5 minutes).
The questions we have for this group are:
How often are you polling user data?
Have you actually benchmarked this to verify that the user is showing as
associated close to the time that you have set for user polling? If so,
what are your polling times for user data, and what is the spread of
times you saw this device as associated?
Is there anyone actually polling their controllers with less than 5
minute polling periods, and have verified performance (connect time to
associated time in Airwave)?
Our test is this: We take a device that was previously known to the
network that has wireless disabled. We locate the device historically
within Airwave (search on the MAC address). We actually click on the
result, and bring up details on that MAC address. We then turn on the
machine’s wireless and start a timer. We keep hitting refresh until we
see the device shown as associated. That is when we record the time. A
few points on this: results are dramatically improved when the load is
low. Right now, we are seeing sub 5 minute times. This is because we
only have 3000 users on the wireless network now. This is not the case
when we have 7000+. We have also been working a lot with Airwave over
this, without any real change in these results. We are wondering if
people just consider this acceptable performance?
Comments/questions/concerns?
Thanks in advance!
-- Jim Gogan
Director, Networking
ITS Communication Technologies
UNC-Chapel Hill
**********
Participation and subscription information for this EDUCAUSE Constituent Group
discussion list can be found at http://www.educause.edu/groups/.
