VERIFIED SOLUTION i

Improving performance for non accumulated batch

Products Affected: EngageOne Server
Product Feature: Batch (Non-Accumulated)
Operating System: AIX
Database: Oracle
 
Results of our performance testing using the supplied template and delivery options.

Test harness specification.

Windows 2008 Server VM, 4 Cores, 8Gb Memory running on a Quad core (8 logical processors, 16Gb memory and SSD)
Benchmarking the virtual drives on the SSD produced >600 Mb/S R/W speeds.
The number of worker threads for the composition phase was set to 8.

The data sets.

Test 1 with 2000 documents of 1 page per document.
This ran in 20 seconds producing a total of 4000 pages @ 200 pps

Test 2 with 2000 documents of 3 pages per document.
This ran in 25 seconds producing a total of 12000 pages @ 480 pps

Test 3 with 2000 documents of 16 pages per document.
This ran in 60 seconds producing a total of 64000 pages @ 1000 pps


The above results are rounded but do suggest a direct correlation between performance and document complexity.


The above tests were repeated using different disk configurations.
A network share was setup that benchmarked at 25Mb/S.

Test 1 ran with the input data and print files on the network share.
This showed a marginal performance drop as the additional overhead was minimal.

Test 2 ran with the batch work directory on the network share.
This showed a three fold decrease in performance when run on the single page data set.
In fact, the results are very similar to those seen by the customer.
This would seem logical as the bulk of the I/O during NA Batch is done through this working directory.

A further test was run to verify the effect of the number of worker threads on performance.
As expected, the number of threads have minimal effect on single page documents.
They do, however, have a more dramatic effect on larger documents.

Conclusion.
The most dramatic performance improvement would be gained by configuring the NA Batch working directory to a local, high performance disk.
If sufficient RAM is available, then a ramdisk would be ideal.

The tests show that the number of pages per document have a substantial influence on performance.
If the average document page count is three and above, then a performance figure approaching 500 PPS can be achieved.
 
UPDATED:  May 29, 2019