[f-nsp] Serveriron 4G with round robin

[Jimmy Stewpot]

Hello,

We are currently using a pair of Serveriron 4G's to load balance mysql on port 3306. We currently have a single master which we use for all write requests and two slave servers, only the slave servers are load balanced to maintain performance in a read-intensive application. The load is evenly distributed between both slaves without any problems. However we want to add an additional server into the group making it 3 servers. The algorithm that we are using for load balancing is round-robin. The problem that we are seeing is when we add the third server into the group all of the load appears to go to that server. What I expected was the following behavior :

Serv1 : 200 conn
Serv2 : 200 conn
Serv3 : 1 conn

Serv1 : 201 conn
Serv2 : 201 conn
Serv3 : 2 conn

What we are actually seeing is that all the connections appear to be going directly to the third server and not being round-robin load balanced as I would have expected. In the documentation I cannot see any mention of the servers being ramped up in a rapid fashion. The configuration looks like this :


!Building configuration...
!Current configuration : 3908 bytes
!
ver 10.2.00dTI4
!
global-protocol-vlan
!

server port 3306
 connection-log all
 tcp
 tcp l4-check-only                                                
!
context default
!
server real Serv1 10.1.1.1
 source-nat                                                       
 port 3306
 port 3306 l4-check-only
!
server real Serv2 10.1.1.2
 source-nat
 port 3306
 port 3306 l4-check-only
!
server real Serv3 10.1.1.3
 port default disable
 disable                                                          
 source-nat
 port 3306 disable
 port 3306 l4-check-only

!
server virtual DBServALL 10.1.1.250
 sym-priority 100
 predictor round-robin
 port default disable
 port 3306
 bind 3306 Serv2 3306 Serv1 3306 Serv3 3306
!

I have tried playing with the slow-start mechanisms without much luck, I tried the following settings.


server port 3306
  slow-start 303 1 60 2 120 240

 connection-log all
 tcp
 tcp l4-check-only                                                
!

server real Serv3 10.1.1.3
 port default disable
 disable                                                          
 source-nat
 port 3306 disable
 port 3306 slow-start 303
 port 3306 l4-check-only
!

But it seemed to have little effect, most of the load went to the third server and thus started causing problems for that system. Has anyone else seen this type of issue in the past?  Does round-robin not really like having uneven server numbers (e.g. 3). Any advice would be really appreciated.

Regards,

Jimmy Stewpot.