(Copied from my other blog at blogs.sun.com. Original post is at: http://blogs.sun.com/mandalika/entry/peoplesoft_north_american_payroll_on)

During the "Sun day" keynote at OOW 09, John Fowler stated that we are #1 in PeopleSoft North American Payroll performance. Later Vince Carbone from our Performance Technologies group went on comparing our benchmark numbers with HP's and IBM's in BestPerf's group blog at Oracle PeopleSoft Payroll (NA) Sun SPARC Enterprise M4000 and Sun Storage F5100 World Record Performance. Meanwhile Jeorg Moellenkamp had been clarifying few things in his blog at App benchmarks, incorrect conclusions and the Sun Storage F5100. Interestingly it all happened while we have no concrete evidence in our hands to show to the outside world. We got our benchmark results validated right before the Oracle OpenWorld, which gave us the ability to speak about it publicly [ and we used it to the extent we could use ]. However Oracle folks were busy with their scheduled tasks for OOW 09 and couldn't work on the benchmark results white paper until now. Finally the white paper with the NA Payroll benchmark results is available on Oracle Applications benchmark web site. Here is the URL:

        PeopleSoft Enterprise Payroll 9.0 using Oracle for Solaris on a Sun SPARC Enterprise M4000

Once again the summary of results is shown below but in a slightly different format. These numbers were extracted from the very first page of the benchmark results white papers where PeopleSoft usually highlights the significance of the results and the actual numbers that they are interested in. The results are sorted by the hourly throughput (payments/hour) in the descending order. The goal is to achieve as much hourly throughput as possible. Since there is one 16 stream result as well in the following table, exercise caution when comparing 8 stream results with 16 stream results. In general, 16 parallel job streams are supposed to yield better throughput when compared to 8 parallel job streams. Hence comparing a 16 stream number with an 8 stream number is not an exact apple-to-apple comparison. It is more like comparing an apple to another apple that is half in size. Click on the link that is underneath the hourly throughput values to open corresponding benchmark result.

Oracle PeopleSoft North American Payroll 9.0 - Number of employees: 240,000 & Number of payments: 360,000

Vendor	OS	Hardware Config	#Job Streams	Elapsed Time (min)	Hourly Throughput Payments per Hour
Sun	Solaris 10 5/09	1x Sun SPARC Enterprise M4000 with 4 x 2.53 GHz SPARC64-VII Quad-Core processors and 32 GB memory 1 x Sun Storage F5100 Flash Array with 40 Flash Modules for data, indexes 1 x Sun Storage J4200 Array for redo logs	8	67.85	318,349
HP	HP-UX	1 x HP Integrity rx6600 with 4 x 1.6 GHz Intel Itanium2 9000 Dual-Core processors and 32 GB memory 1 x HP StorageWorks EVA 8100	16	68.07	317,320
HP	HP-UX	1 x HP Integrity rx6600 with 4 x 1.6 GHz Intel Itanium2 9000 Dual-Core processors and 32 GB memory 1 x HP StorageWorks EVA 8100	8	89.77	240,615*
IBM	z/OS	1 x IBM zSeries 990 model 2084-B16 with 313 Feature with 6 x IBM z990 Gen1 processors (populated: 13, used: 6) and 32 GB memory 1 x IBM TotalStorage DS8300 with dual 4-way processors	8	91.7	235,551

This is all public information -- so, feel free to draw your own conclusions. ^*At this time of writing, HP's 8 stream results were pulled out of Oracle Applications benchmark web site for some reason I do not know why. Hopefully it will show up again on the same web site soon. If it doesn't re-appear even after a month, probably we can simply assume that the result is withdrawn.

As these benchmark results were already discussed by different people in different blogs, I have nothing much to add. The only thing that I want to highlight is that this particular workload is moderately CPU intensive, but very I/O bound. Hence the better the I/O sub-system, the better the performance. Vince provided an insight on Why Sun Storage F5100 is a good option for this workload, while Jignesh Shah from our ISV-Engineering organization focused on the performance of this benchmark workload with F20 PCIe Card.

Also when dealing with NA Payroll, it is very unlikely to achieve a nice out-of-the-box performance. It requires a lot of database tuning too. As the data sets are very large, we partitioned the data in some of the very hot objects and it showed good improvement in query response times. So if you are a PeopleSoft customer running Payroll application with millions of rows of non-partitioned data, consider partitioning the data. We are currently working on a best practices blueprint document for PeopleSoft North American Payroll that presents a variety of tuning tips like these in addition to the recommended practices for F5100 flash array and flash accelerator F20 PCIe card. Stay tuned ..

Related Blog Post:

• PeopleSoft HRMS 8.9 Self-Service Benchmark on M3000 & T5120 Servers