03-28-2010 03:19 PM
Hi All
I was wondering if any one could help with the following, I have an ACE module running 2.2 software and would like to change the predictor to least connections and add sticky for this,
-------snippet-------
probe http http-default
interval 10
passdetect interval 30
expect status 200 200
connection term forced
rserver host one
ip address x.x.x.x
inservice
rserver host two
ip address x.x.x.x
inservice
rserver host three
ip address x.x.x.x
inservice
rserver host four
ip address x.x.x.x
inservice
serverfarm host serverfarm
predictor hash address source
probe vle-http-probe
rserver one
rserver two
inservice
rserver three
inservice
rserver four
inservice
class-map match-all VL10-class
2 match virtual-address x.x.x.x any
class-map match-all VL-class
2 match virtual-address x.x.x.x any
class-map type management match-any remote_access
2 match protocol ssh source-address x.x.x.x
3 match protocol telnet source-address x.x.x.x
4 match protocol icmp any
policy-map type management first-match remote_mgmt_allow_policy
class remote_access
permit
policy-map type loadbalance first-match Virt_Policy
class class-default
serverfarm serverfarm
policy-map multi-match bv-policy
class VL-class
loadbalance vip inservice
loadbalance policy Virt_Policy
loadbalance vip icmp-reply
interface vlan 100
access-group input ALL
service-policy input remote_mgmt_allow_policy
service-policy input bv-policy
no shutdown
Any help would be very much appreciated.
Regards MJ
03-29-2010 04:47 AM
Here is what you need to do :
serverfarm host serverfarm
predictor leastconn slowstart 60
stick ip-netmask 255.255.255.255 address source STKY-SRC-IP
serverfarm serverfarm
policy-map type loadbalance first-match Virt_Policy
class class-default
sticky-serverfarm STKY-SRC-IP
However, I'm affraid you won't see much difference with your actual config.
You probably see more traffic on one rserver and are trying to correct this with predictor leastconn but still keeping stickyness.
Stickyness by its nature breaks any "predictory" algorithm. So don't expect to see a better load distribution.
Gilles.
03-29-2010 08:07 AM
Hi Gilles
Thanks for the response, yes that is the problem I am seeing... Is there any way around this? Will any of the predicto's help on this?
Many thanks MJ
03-29-2010 10:08 AM
As I said, because of stickyness no predictor method will solve your problem.
For example, take 2 usrs.
First one opening 10 connections and 2nd one opening 1 connection.
Because of stickyness the first server gets 10 connections and the other server only 1.
If you try to split the load evenly, you break stickyness.
Also, because there are mega proxy on the internet (many users behind a single ip) doing sticky source ip can be dramatic.
If this is HTTP traffic, you should try to use sticky based on cookie.
gilles.
03-30-2010 07:48 AM
Hi Gilles
Thanks for your hekp on this..... looking at the configuratuion guide the following would need to be added for sticky based on cookie's....
serverfarm host serverfarm
predictor leastconn slowstart 60
sticky http-cookie app-cookie
cookie insert
serverfarm serverfarm
policy-map type loadbalance first-match Virt_Policy
class class-default
sticky-serverfarm app-cookie
Regards MJ
03-30-2010 09:06 AM
This is correct.
Except for the sticky definition.
You need to give a name to the cookie and a name to the sticky group.
This can be the same name.
AceC6k2/Admin(config)# sticky http-cookie app-cookie ?
Gilles.
03-31-2010 02:50 AM
Hi Gilles
Thanks for your help on this, I take it the sticky group name is not referenced any where else in the configuration then.
Regards MJ
Discover and save your favorite ideas. Come back to expert answers, step-by-step guides, recent topics, and more.
New here? Get started with these tips. How to use Community New member guide