Human Factors of XR: Using Human Factors to Design XR Systems
Oracle Retail Merchandising System 13.1.1 performance on HP BladeSystem
1. Performance brief for Oracle Retail
Merchandising System 13.1.1 on HP
ProLiant BL460c G6 and EVA6400
Using Oracle Database 11gR1 RAC on Linux
Technical white paper
Table of contents
Executive summary............................................................................................................................... 2
Introduction ......................................................................................................................................... 2
Test topology....................................................................................................................................... 2
Test methodology................................................................................................................................. 4
Test business processes......................................................................................................................... 5
Online processes ............................................................................................................................. 5
Batch processes ............................................................................................................................... 5
Test results .......................................................................................................................................... 6
Test analysis summary .......................................................................................................................... 6
Online ............................................................................................................................................ 6
Batch .............................................................................................................................................. 6
Appendix A – Benchmark environment ................................................................................................... 9
For more information .......................................................................................................................... 10
2. Executive summary
Oracle® Retail Merchandising System (RMS) load tests for both batch and online are run on an
Oracle Real Application Clusters (RAC) database in an HP BladeSystem environment. As nodes are
added to the RAC database, run times are reduced for the batch workloads and CPU utilization is
reduced for equal numbers of online users. Online response times are excellent for all configurations
with little variation as RAC nodes are added to the configuration. For a batch workload, the
performance increase in going from 1 to 2 and 2 to 4 nodes is typically around 1½ times. For a
consistent online workload, CPU utilization goes down to 55 – 60%.
This white paper describes only the relative performance of RMS 13.1.1 as RAC nodes are added to
the database. Detailed test results are available in a performance report available from the HP
solutions alliance engineer for Oracle Retail. Refer to the URL in the For more information section of
this brief.
This paper describes testing completed in April 2010.
Introduction
In April 2010, the HP solution alliance engineering team for Oracle completed a performance test
using the Oracle Retail Merchandising System (RMS) 13.1.1 Benchmark Kit on Oracle Database
11gR1 Real Application Clusters (RAC). The performance test measured the online and batch
performance of the RMS application in multiple RAC configurations to determine the horizontal
scalability of RMS 13.1 on HP ProLiant BL460c G6 servers running Oracle Enterprise Linux (OEL)
release 5 update 2.
The online tests consisted of RMS functionality that represents a day in the life of an RMS user. Tests
included twelve workflows spanning numerous functional areas such as cost adjustments,
organizational hierarchy, deals, items, mass return transfers, orders, return to vendors, suppliers and
transfers. The batch tests included two workflows; one file-based batch process, sales upload, and the
other table-based batch process, replenishment.
These test results can be used to understand RMS 13.1.1 scaling out on an Oracle RAC database.
The environment was set up for RAC scalability testing only. For a highly available environment,
another application server could be added to eliminate that single point of failure. A load balancer
could also be utilized to distribute online requests between the application servers.
Test topology
Oracle Retail Merchandising System 13.1.1 was run on HP BladeSystem ProLiant servers with an HP
StorageWorks 6400 Enterprise Virtual Array for storage as shown in Figure 1.
2
4. The RMS database was run on 1, 2, 3 and 4 RAC nodes (with host names rac1, rac2, rac3, and
rac4) running Oracle Database 11.1.0.7 on ProLiant BL460c G6 blade servers with Oracle
Enterprise Linux 5 update 2 (OEL 5u2) within a c7000 blade enclosure. Each BL460c server was
populated with a single quad-core 2.93GHz Intel® Xeon® processor and 24 GB of memory. For
greater loads, these blade servers can be configured with two quad-core processors and up to 192
GB of memory. A Virtual Connect Flex-10 Ethernet module provided a 10GbE private interconnect
between the nodes as well as a 1GbE connection outside of the enclosure and public network to the
application server. For 10GbE data center environments, the Flex-10 modules allows up to 10Gb for
the uplink outside the enclosure.
For a baseline comparison, the RMS database was also run on a non-RAC database with host name
rac0 on one of the BL460c G6 blade servers. A Virtual Connect server profile for each database
server was created and assigned as required to the blade servers.
For online tests, a ProLiant BL685c G1 blade server with 4 dual-core 2.6GHz AMD Opteron™
processors and 64 GB of memory running OEL 5u2 in the c7000 enclosure served as the application
server running Oracle Application Server 10.1.2.3. Online user transactions were generated by HP
LoadRunner 9.50 running on a ProLiant DL380 G3 with Microsoft® Windows® Server 2003
connected to the c7000 enclosure through an HP ProCurve 2848 switch.
All disk storage was provided by an HP StorageWorks 6400 Enterprise Virtual Array with dual
HSV400 controllers and 48 300GB 15K RPM SAS drives in 4 enclosures. Each of the blade servers
utilized two fiber channel interfaces with each connected to a separate Virtual Connect 4Gb Fibre
Channel (VC-FC) module. Two fibre channel interfaces from each of the VC-FC modules were
attached to the HP StorageWorks SAN 4/16 switch and the 4 ports from each of the HSV400
controllers were connected to the SAN switch for maximum storage throughput. All of the blade
servers used Virtual Connect fibre channel technology to boot from EVA LUNs. All database files
resided in Oracle Automatic Storage Management (ASM) managed EVA LUNs.
Details of the environment can be found in Appendix A.
Test methodology
The online workload test was conducted using HP LoadRunner software to simulate concurrent users
within the Oracle Retail Merchandising System. The test execution consisted of 30 minutes of user
ramp up followed by 1 hour of steady state. Measurements were recorded across both the
application and database servers during the tests.
The Sales Upload and Replenishment batch workloads were executed via a custom test harness script
running on node rac1. The script controlled process execution, data preparation, file movement and
statistics gathering.
To test performance of a single RAC node, an Oracle database service was created to run against a
single instance of a 2 node RAC cluster and tnsnames.ora designated this service for the RMS
database.
All system performance information was gathered at 15 second intervals by HP Performance Agent
5.00.
4
5. Test business processes
Online processes
The online workload spanned numerous functional areas within RMS. Online tests were run to
simulate the work of 600 simultaneous users. The online transactions that make up each functional
area for the online load test are described below.
Cost Adjustment Create an average cost adjustment for an item.
Warehouse Create a warehouse within the organizational hierarchy.
Deals Search for an approved supplier deal.
Item Creation Create and approve a one-level item with multiple suppliers, stores, multiple
UDAs and various seasons.
Item Creation 2 Create and approve a two-level item with differentiators, multiple suppliers
and a location list of stores.
Mass Return Transfer Create a mass return transfer for an item at all stores.
Order Creation Create and approve a one-level item with multiple suppliers, stores, multiple
UDAs and various seasons.
Order Creation 2 Create an order for a parent item and item differentiators and distribute to
two locations.
Return to Vendor Create a return to vendor for a particular item and location.
Stores Create a store within the organizational hierarchy.
Supplier Create a new supplier within the merchandising system.
Transfer Create a manual requisition transfer of a single item between two locations.
Batch processes
The Point of Sales Upload (posupld) module processes sales and return details from an external point
of sale system. The sales/return transactions are validated against Oracle Retail item/store relations
to ensure the sale is valid, but this validation process can be eliminated if the sales that are being
passed in are screened by sales auditing (Oracle Retail Sales Audit). Sales audit was turned off for
the testing of this program. Three different runs with uploads from 250, 500, and 1000 files were
executed for the non-RAC and each RAC configuration.
Replenishment allows the customer to automate the ordering process for items by monitoring inventory
conditions and creating orders and/or transfers based on predefined replenishment parameters. The
replenishment workload consists of a series of batch modules, some of which require pre- and post-
processing as indicated:
rplatupd (pre, post) Replenishment Attribute Update
rilmaint (post) Replenishment Item Location Maintenance
ociroq (pre) Recommended Order Quantity
reqext (post) Replenishment Quantity Extract
rplext (pre, post) Vendor Replenishment Extraction
rplbld Replenishment Order Build
rplapprv (pre) Automatic Replenishment Order Approval
Including pre and post processing, there are fifteen separate modules. Three different runs with 1.34
million, 2.7 million, and 5.4 million item/locations were executed for the non-RAC and each RAC
configuration.
5
6. Test results
Detailed test results are available from the HP solutions alliance engineer for Oracle Retail. Refer to
the URL in the For more information section of this brief.
Test analysis summary
Online
Online response times were excellent for all configurations with little variation as RAC nodes were
added to the configuration. Since 600 users were run in all configurations rather than running the
maximum number of users that could be accommodated, a comparison of CPU utilization on the
database servers is the best way to compare the scaling of RMS online across RAC nodes. Since
each connection uses some memory, a comparison of memory utilization is also useful to see how
much memory utilization is reduced on each node as more nodes are added.
Figure 2 shows the reduction in CPU and memory utilization as RAC nodes are added to the
configuration. Using CPU as a measure of scaling, we see 1.8x scaling when increasing from 1 to 2
RAC nodes and 1.7x scaling when increasing from 2 to 4 RAC nodes.
Figure 2. Comparison of online scaling in non-RAC and 1- to 4-node RAC configurations
RMS 13.1 Online Scaling on RAC
non-RAC
%CPU/GB
1 RAC node
2 RAC nodes
3 RAC nodes
4 RAC nodes
Avg CPU util (%) Peak Mem (GB)
Batch
A graphic comparison of batch run times for largest loads (1000 posupld files, 5.4 million
replenishment items/locations) in Figure 3 shows the relative performance increase as more nodes are
added to the RAC cluster.
6
7. Figure 3. Comparison of batch scaling in non-RAC and 1- to 4-node RAC configurations
RMS 13.1 Batch Scaling on RAC
non-RAC
Elapsed time
1 RAC node
2 RAC nodes
3 RAC nodes
4 RAC nodes
Replenishment POS Upload
POS upload scaling from 1 to 2 RAC nodes was 1.14x and from 2 to 4 RAC nodes was 1.54x.
Overall replenishment scaling from 1 to 2 RAC nodes was 1.55x. Scaling from 2 to 4 RAC nodes
was 1.54x.
Some replenishment modules ran slower as RAC nodes were added. By tuning the test to have just
those modules run against a single RAC node while the other modules ran across all RAC nodes,
overall horizontal scaling for replenishment was improved. Scaling of individual replenishment
module run times are shown in Figure 4.
7
8. Figure 4. Comparison of individual replenishment module scaling in non-RAC and 1- to 4-node RAC configurations. Modules
that completed in less than ten seconds not included.
RMS 13 Replenishment Module
Scaling on RAC
Elapsed time
rplatupd
rilmaint
ociroq
reqext
rplext
rplbld
0 1 2 3 4
Number of RAC nodes (0=non-RAC)
8
9. Appendix A – Benchmark environment
Database Servers
Four HP ProLiant BL460c blade servers were used as database and batch servers. Each system was
configured as follows:
Processor: One quad-core 2.93GHz Xeon 5570 with hyper-threading enabled
Memory: 24 GB
Network: Embedded NC532i Dual Port Flex-10 10GbE Multifunction Server Adapter
Storage HBA: QLogic QMH2462 4Gb FC HBA
Cluster Interconnect: 10GbE with jumbo frames
Operating System: Oracle Enterprise Linux Release 5 Update 2 – 64-bit
Storage
HP StorageWorks 6400 Enterprise Virtual Array
2 x HSV400 controllers
48 x 300GB 15K RPM SAS drives in 4 drive bays
Application Server
One HP ProLiant BL685c G1 blade server was used as the applications server for all online tests. It
was configured as follows:
Processor: Four dual-core 2.6GHz AMD Opteron 8218
Memory: 64 GB
Network: Embedded NC373i Multifunction Gigabit Server Adapter
Operating System: Oracle Enterprise Linux Release 5 Update 2 – 64-bit
Software Versions
Oracle Retail Merchandising System 13.1.1
Oracle Database 11gR1 11.1.0.7 with patches:
7697360
7378322
7036284
7272646
Oracle Application Server and Developer Suite 10gR2 10.1.2.3
Oracle Application Server Forms and Reports Services 10g 10.1.2.0.2 with patch:
7379122
HP LoadRunner 9.5
HP Performance Agent 5.0
Oracle Cluster File System OCFS2 1.2.9
9