IBM Mainframe Forum

[Antonyraj85]

Hi friends,

Is it possible to have a more than 100 service classes in the single WLM policy. It is required for my Env. Do you have any idea. Please share with me.

Regards,
Antony

[Robert Sample]

The manual says 100 is the maximum. There is a consistent rule of thumb that I've heard many times over the years that having more than 30 services classes can be detrimental to system performance. This is because WLM can only adjust performance for one service class per 10-second interval; having 100 services classes could mean more than 15 minutes between adjustments to a service class. If the PI is way off for that service class, such a long time between adjustments could seriously impact the system.

[dick scherrer]

Hello,

Why does someone believe more than 100 is needed? This sounds like there is a considerable misunderstanding. . .

[Antonyraj85]

Hi Dick,

We have 100s DB2 subsystem. So we wants to generate db2 performance report based on the each subsystem(Thru report class). So we need to create separate service class for each. The LPAR is basically for TEST and Development purpose. So can you please suggest how we can overcome this issue. Is there any alternate ?

Regards,
Antony

[Robert Sample]

You're really not making a lot of sense. If you have 100 development and 100 QA DB2 subsystems, you should manage them as two service classes -- one for development and one for QA. The whole point of WLM is that you group similar work into service classes, prioritize the service classes, and manage the system as a whole. WLM is not designed to manage resources in the way you want to. Furthermore, if you have multiple LPARs in your sysplex, all of them are managed through one WLM policy -- so you could potentially be impacting a lot of work through your quest for the impossible.

If you really decide you want performance data on each individual DB2 subsystem, that is what SMF data is for. If your site uses MXG, I'd start with the ASUMCEC and RMFINTRV reports to get a feel for what's happening overall with the system, then dig into anomalies on a case by case basis.

However, you simply cannot manage "hundreds" of subsystems on a one-by-one basis. Your system overhead for WLM will go through the roof, workloads won't meet goals, WLM will be very slow to correct issues, and you'll be lost in a flood of data that is too detailed to make sense of.

[Antonyraj85]

Hi Robert,

Thanks a lot.

Regards,
Antony