Forgive me if this is the wrong forum but it was the closest one I found relating to my issue.
I've trying to load balance four of our radius servers using IOS SLB. The config works well and the radius servers are accepting requests fine. I follow this article which wasn't too bad to follow:
http://www.cisco.com/application/pdf/en/us/guest/netsol/ns377/c649/cdccont_0900aecd800eb95f.pdf
My two questions are:
1. Sticky Option
I understand it's used to make sure the client's accounting information
goes to the correct real server, but I'm not sure how it really works
and what's the best time to set it to.
Eg:
ip slb vserver RAD-UDP-1646
virtual 210.x.x.224 udp 1646
serverfarm RADFARM
sticky 86400 group 10
inservice
a/ The documentation says "This configuraion causes the sticky database to store its entries for 86,400 seconds of inactivity". What do they mean by "inactivity" - no radius packets coming through? inactivity from the user's end?
b/ It also says "the client's IP address is added to the IOS SLB database..." - is this the client's framed IP that the ISP assigns to the customer???
c/ And what would be the optimum time to set the sticky timer to be?
2. SLB connection statistics
core1-router#sh ip slb reals
real farm name weight state conns
-------------------------------------------------------------------
203.x.x.74 RADFARM 8 OPERATIONAL 0
203.x.x.78 RADFARM 8 OPERATIONAL 0
203.x.x.79 RADFARM 8 OPERATIONAL 0
203.x.x.80 RADFARM 8 OPERATIONAL 2
When you disconnect, the slb stats still show you as being connected to
the real server (and both udp ports) which isn't very accurate. There is a default "delay" time which handles TCP disconnections and after being disconnected for 10 sec, the SLB stats are updated to reflect this (I've verified this works)- but nothing about how it handles UDP disconnections??? This
would skew the stats and give us a very bad misrepresentation of the
number of current and valid connections. Is there anyway to correct this???
Thanks.
Andy
