SUN/ORACLE BENCHMARK STUDY
About DeVry University
Founded in 1931 and headquartered in Oakbrook Terrace, IL, DeVry University provides career-oriented undergraduate and graduate programs in business, technology, and management to a diverse and geographically dispersed population of almost 50,000 students. With more than 70 locations in the United States and Canada, including online delivery of programs, DeVry is one of the largest private higher-education systems in North America.
DeVry University runs the first and the largest Oracle Student System (OSS) implementation in the world, serving the university's faculty, administrative staff, and students. The OSS implementation extends to all of the 70 active business operations locations, one third of which have their own networks. In addition, tens of thousands of users access the system over the Internet via a Web-based interface.
About the Oracle Student System
The Oracle Student System (OSS) is a comprehensive software solution for institutions of higher education. OSS provides an integrated suite of software for managing student recruiting, admissions, and enrollment; program structure and planning; and academic records. OSS is based on Oracle's E-Business Suite (AKA Oracle Applications 11i) and uses the Oracle Database to manage a data store containing information about students, faculty, staff, recruits, classes, grades, and other data.
DeVry's OSS Implementation
DeVry's OSS implementation is housed in a single, central data center. The application is deployed on the Sun platform using mid-range Sun Fire servers (Sun Fire E6800 servers with 24 CPUs, Sun Fire 4800 servers, several workgroup servers), the Sun StorEdge 9960 storage system, and the Sun Solaris 8 operating system. The DeVry network is powered by Cisco and Nokia equipment.
DeVry had encountered two main challenges in its OSS deployment:
- Performance slowdowns during peak loads, especially during class registration periods
- An inability to adequately predict the required system resources due to a lack of information about scaling to meet varying load volumes
Traffic spikes occur at the beginning and end of each academic term as students register for classes. During heavy loads, the system administrator can help performance by deferring transaction and batch-oriented activities to off-peak times. Still, as is typical in institutions of higher education, the infrastructure for the OSS is fully utilized at only a fraction of the time (perhaps only 10%), while much of the system remains idle during off-peak times.
Performance Tuning at the Sun Benchmark Center
Sun Microsystems invited DeVry University and its vendors to conduct a performance tuning project at the Sun Benchmark Center in Newark, California. Utilizing the entire Sun product line, these state-of-the-art facilities provide technical expertise for large, scaleable, high performance customer benchmarks. There are six Sun Benchmark Centers around the world:
- Frankfurt, Germany, center for EMEA benchmarks
- Hillsboro, OR, specializing in HPC benchmarks
- Manchester, England, center for EMEA benchmarks
- Newark, CA, specializing in ecommerce and database benchmarks
- Paris, France, center for EMEA benchmarks
- Tokyo, Japan, center for Asia-Pacific benchmarks
The benchmark study was designed to test DeVry's OSS implementation for performance and scalability, and to provide a stable and scalable OSS environment. The overall goal was to determine how to double the performance of the OSS implementation (reducing transaction and screen paint times by half) and, where possible, simplify the implementation, in the DeVry environment.
In addition to Sun, participating vendors included:
- Cisco — network infrastructure and load balancers
- Fujitsu — firewalls
- Mercury Interactive — system test applications
- Utopia Solutions — system test expertise
- Oracle — Student System Software
Environment Replication
The plan for the benchmark study was to replicate, as closely as possible, DeVry's production environment in the Sun Benchmark Center. The test environment replicated the key elements of the DeVry production environment:
- Network infrastructure
- Hardware
- Software, including an identical implementation of the OSS release IGS.K that was then running at DeVry, and IGS.L (which DeVry was then evaluating)
- Actual production data (200GB copied from DeVry's OSS database)
Load Simulation
In addition to replicating the DeVry production environment, test scripts were designed to generate business transaction loads by simulating typical user activities (navigating the user interface and entering data) in the DeVry OSS implementation. These test scripts were written by Utopia Solutions in the Mercury Interactive LoadRunner language and designed to test key functional areas of the OSS application.
In addition to replicating the DeVry production environment, test scripts were designed to generate business transaction loads by simulating typical user activities (navigating the user interface and entering data) in the DeVry OSS implementation. These test scripts were written by Utopia Solutions in the Mercury Interactive LoadRunner language and designed to test key functional areas of the OSS application.
By running workload simulation tests and observing the system under varying transaction loads (based on the number of concurrent users), the participants were able to collaborate on tuning the configuration to maximize performance, throughput, and scalability. Once the baseline configuration was established and tested, participants could run the performance tests iteratively to determine the performance and throughput effect of various configuration changes.
The test scripts generated the following business transactions:
- Enrollment (business transaction: enroll a student)
- Grading (business transaction: manage grades)
- Admissions (business transaction: enter an application)
Results
To begin, load testing was conducted on the default configuration to obtain baseline performance data. The participants were able to use this baseline data for comparison purposes when subsequently making configuration changes and collecting performance data.
Tuning the Load Balancer — In the DeVry environment, when users logged into OSS, they were assigned to one of only two application servers. This approach resulted in a load bottleneck in the two application servers. The participants tuned the load-balancing algorithm to fully utilize all four application servers, which resulted in an approximate overall improvement of 40% in transaction response times.
Moving the Concurrent Manager to the Database Server — Moving the Concurrent Manager from the application server tier to the database server also yielded performance improvements. The number of disk operations per second dropped significantly, and the database CPU utilization dropped 30%. These improvements enabled the testers to ramp up the number of users.
Moving Application Server Machines to a Non-Secured Location — The Application Server machines were moved to the DMZ (non-secured zone in the network) to determine whether removing the overhead of secure connections had a measurable performance impact. This changed proved to show little, if any, performance improvement.
Based on this study, DeVry revised its deployment architecture to handle load balancing more efficiently, resulting in substantial performance gains in their OSS implementation. Because the DeVry infrastructure was re-created at the Sun facility, many proposed changes could be tested without fear of impact to the live production environment.
To read the complete Sun/Oracle benchmark study at DeVry University, click here.
For additional information on the Sun Benchmark Center, visit http://www.sun.com/visitors-info/benchmarks/, contact education_news@sun.com or click here to have your local Sun representative contact you.
| 
