ADABAS Performance HP-UX i11.3

, , , ,
1

I have read some great threads on this discussion community.

In the ‘Search for performance’ the different areas and ‘trial and error’ methodology, I have found that ‘tweaking’ must occur in many places. ONLINE vs Batch, Message Brokers and Replication bring to the table more and more ‘tweaking’ either because of problems or the search for better performance.

One thing I am not sure of.
If I have 8 cpus I use the calculation CPU * 2 for threads in ADABAS. But in some conversations concerning buffer flushes, it was recommended increasing threads with the other parameters like write-limit, bfio, etc.

One of the things I noticed in performance tuning, ‘levling’ Unix Message Queues seems to ‘calm’ the system down. We have seen CPS go from an average of 500 to 2,000 after leveling and buffer tweaking.

If you increase the number of threads, do you not incur overhead with CPU ‘marshaling’, memory locks, etc?

Thanks in advance.

Mike Waldron

2

I think it was Adabas 7 for mainframes where the CPU consumption for increasing threads was no longer much of a concern. I assume some of the same performance enhancements have been made in LUW versions as well?

While we still don’t want to severely overdo NT, it seems to be much better to overallocate NT than risk having commands in the last thread.

3

Thanks Brian. The only thing ADABAS Mainframe has in common with Unix ADABAS is the name.
Threads are not the same. You will see an equal amount of commands in each thread. Equal distribution.

Mike

4

I think I may have found something. Complex searches are been done fairly often.
The latest release of ADABAS Open Systems has a performance enhancement where complex searches will not consume all the threads and are limited to a set of threads. Prior to this, users had to wait sometimes if a complex search was being done.

Will let you know what I find out.

Thanks

Mike W.