Oracle 11g versus Oracle 10g

Edit online

During the test phase the new Oracle 11g version became available for Linux on System z. This offered the possibility to compare Oracle 10g with Oracle 11g.

User scaling and transactional throughput

Figure 1 shows the user scaling results with Oracle 10g as shown in Figure 1 with the results from Oracle 11g for the transactional throughput.

Figure 1. Oracle 11g versus Oracle 10g, user scaling and transactional throughput
Graph showing user scaling and transactional throughput

Observation

The transactional throughput is increased by more than a factor of 4.

Conclusion

This is a very impressive improvement for the new version, and leads to the question, what causes this improvement?

User scaling and disk throughput

To get a better understanding what has changed in the behavior with Oracle 11g, Figure 2 shows the disk I/O behavior for version 10g and 11g.

Figure 2. Oracle 11g versus Oracle 10g, user scaling and disk throughput
graph showing user scaling and disk throughput

Observation

The disk read rates has dramatically decreased, around -90%, while the write rates have increased between a factor of 3 and 4.

Conclusion

The lower read rate is a clear indicator, that the caching is improved in terms of efficiency. The database now finds much more data in the caches and needs to read less data from disk. The increase in write rates follows roughly the throughput increase. This confirms that our transactional throughput has really improved that much.

User scaling and CPU cost per transactional throughput

Having more of the required data in the caches reduces the waits and means that the CPUs can drive a greater workload. This is of cause related to a higher CPU load, therefore the CPU cost to drive the workload it is much more interesting. This is measured in units of how many transactions are driven by one CPU. A higher value indicates that the cost is lower, that means less CPU is required for the same amount of workload. The normalized values are shown in Figure 3.

Figure 3. Oracle 11g versus Oracle 10g, user scaling and CPU cost per transactional throughput
graph showing user scaling and CPU cost per transactional throughput

Observation

The amount of transactions driven per CPU is increased between 45% and 70%. The system CPU is very similar, and the major contribution comes from the user space CPU.

Conclusion

The results show that we have not only a dramatic improvement in transactional throughput, the CPU effort per transaction is also lower, which highly recommends the update to Oracle 11g, at least for transactional workloads.