ACE isn't load balancing as expected

[gavin han]

Hi,

I've issue with two servers being load balanced on ACE. traffice isn't distributed equally:

probe tcp ProbePort80

  port 80

  interval 15

  passdetect interval 60

  open 1

rserver host Server1

  ip address 10.10.10.1

  probe ProbePort80

  inservice

rserver host Server2

  ip address 10.10.10.2

  probe ProbePort80

  inservice

serverfarm host Server-1-2-Loadbalanced

  failaction purge

  rserver Server1

    inservice

  rserver Server2

    inservice

sticky ip-netmask 255.255.255.128 address both Server-1-2-Loadbalanced

  timeout 60

  timeout activeconns

  replicate sticky

  serverfarm Server-1-2-Loadbalanced

class-map match-any Server-1-2-Loadbalanced

  2 match virtual-address 192.168.1.150 tcp any

policy-map type loadbalance first-match Server-1-2-Loadbalanced

  class class-default

    sticky-serverfarm Server-1-2-Loadbalanced

policy-map multi-match Load_Balanced_Vlan9

  class Server-1-2-Loadbalanced

    loadbalance vip inservice

    loadbalance policy Server-1-2-Loadbalanced

    loadbalance vip icmp-reply

ACE# sh serverf Server-1-2-Loadbalanced

serverfarm     : Server-1-2-Loadbalanced, type: HOST

total rservers : 2

---------------------------------

                                                ----------connections-----------

       real                  weight state        current    total      failures

   ---+---------------------+------+------------+----------+----------+---------

   rserver: Server1

       10.10.10.1:0      8      OPERATIONAL  134        387553153  76428027

   rserver: Server1

       10.10.10.2:0      8      OPERATIONAL  3270       192367801  32921205

as you can see above. 1st server got only 134 connections while 2nd server got 3270.

could help please.

[Surya ARBY]

Change the predictor and see if it helps :

http://www.cisco.com/en/US/docs/app_ntwk_services/data_center_app_services/ace_appliances/vA4_2_0/configuration/slb/guide/rsfarms.html#wp1000073

[pablo.nxh]

Hi Gavin,

You may need to change the sticky netmask from /25 to /32 as shown in your output all the traffic from a single subnet is being stuck to the same real server.

sticky ip-netmask 255.255.255.255 address source Server-1-2-Loadbalanced

HTH

__ __

Pablo

[gavin han]

Hi, thanks for your reply. could you advise the best practice to use stick netmask? how do we know what sticky netmask to use?

thanks.

[Surya ARBY]

Hi.

It depends on the requirements of your environment : using a /32 mask will provide better LB granularity (each incoming IP address will be load balanced as a new entity) but it will take a stick entry in the table for each client ip address.

Using larger masks, you'll lose LB granularity (LB will not be done per ip address but per network instead) but you'll save space into the sticky table, keeping it very small with large masks.