Scaling DB2 Connect MAXAGENTS

In these tests, we measured the affects of connection concentration at DB2 Connect™. Two virtual CPUs were defined to the DB2 Connect z/VM® guest and 1 GB of memory was used. The number of user threads was held constant at 500. The z/VM LPAR was assigned six CPUs and the z/OS® LPAR was assigned four CPUs. The MAXAGENTS were varied from 50 to 500 and think time values of 2500 and 5000 milliseconds were used.

Figure 1 shows throughput for a DB2 Connect z/VM guest with think times of 2500 and 5000 and scaling MAXAGENTS from 50 to 500.

Figure 2 shows CPU utilization for DB2 Connect z/VM guest with think times of 2500 to 5000 milliseconds and scaling MAXAGENTS from 50 to 500.

Observations

Maximum throughput is reached at 100 MAXAGENTS in both the 2500 and 5000 milliseconds think time cases. More MAXAGENTS are not required for this workload. In the 2500 milliseconds case there is a significant increase in throughput at 100 MAXAGENTS, compared to 50 users.

Conclusion

For the IRWW workload, a MAXAGENTS value of 100 is sufficient for this workload, even though the number of users is 500. This demonstrates that limiting the connections to DB2® on z/OS can be done without degrading the throughput, up to a certain load level. It can also be used to limit the load created from this source on DB2 to a certain level, protecting the database against load peaks from that source.