SMLG load balance

[Former Member]

Dear Experts,

We are using logon groups for load balance. Logon quality is used to determine the next logon server.

We want to change this algorith with something like round robin. Each dialog instance should accept one user so user count will be identical on all dialog instances.

Any way to perform this?

best Regards

[Former Member]

This message was moderated.

[Former Member]

thanks for response.

We want roung robin mechanism during dialog logon.

As I see Note 593058 - New RFC load balancing procedure

describes actions for RFC connections for logon group.

Dees this method work for dialog logons. We tried the method but not worked.

[Former Member]

As far as I know, the process I posted above works for dialog logins as well.

[Former Member]

To change logon balancing to Round Robin, follow these steps:

1. In SMLG, double click the group and under attributes tab check "Ext. RFC-enabled"

2. Goto SE16 and modify table RZLLICLASS.

3. Select for the group (CLASSNAME) from step 1, change FAVTYPE from "B" to "R"

[rkresha]

3. Select for the group (CLASSNAME) from step 1, change FAVTYPE from "B" to "R"

Should be from "B" to "W"

[Former Member]

Dear Sez .

I am not sure about changing the algorithm to round robin as desired but we can improve poor distibution of user.

++++++++++++

RSRZLLG0 is the report that updates the data used by SMLG. RSRZLLG0

itself is called by SAPMSSY6 which is an automatically started report (aso-called 'autoabap', which will be started all five minutes. Or, to be

more exactly, will be started at the interval that is given in the

parameter rdisp/autoabaptime (default 300 seconds).

Please go to transaction SMLG, and than goto -> load distribution.

There you will see a response time of the individual application

servers.

1. Assign Heavy transaction user groups to the servers with more

computing power and memory. This will ensure that even if these users use the resource concurrently, they will not face performance issues

to a greater extent.

2. Estimating the optimum value for response time and logged in users is Difficult. But you can definitely base these values on the average

response of the server that is acceptable to the users and they can

work comfortably without performance issues. The number of users can

be based upon the situations that you have faced when :

a) n Number of users were logged in and the performance of the server was acceptable.

b) y Number of users were logged in and users started complaining

about the server response.

The limits that we discussed do influence the algorithm for load

balancing to some extent. But at the same time we cannot fully influencethe same. However you can try to implement the changes and further fine

tune the limits as per requirements.

Due to the large fluctuations in response times for each instance

then you should consider the changes in the logon load balancing

algorithm of note #51789 (Poor user distribution in logon distribution)

to give more weighting to the users than response time than default

for Quality values in SMLG to get more equal load balancing between

the instances.

Hope this information may / may not help you

Rgds