Benchmark Using a Transaction Accelerator
A benchmark, conducted by IBM's High Volume Websites (HVWB) Lab, compared the performance of IBM's WebSphere Application Server and DB2 with and without Versant's enJin product.1 The benchmark is a stock trading application with simple and complex transactions.
Versant no longer markets enJin. Nevertheless, this benchmark is still valid and illustrates the performance gain when using an object database in the middle tier. Versant called its enJin product a "transaction accelerator." It worked with Java application servers in much the same way as described in the other architectures that you can find in the related content below. The enJin product was built upon Versant's object database management system.
The performance gains possible with an object database in the middle tier are illustrated in the following graphs. They compare performance with and without enJin in the J2EE architecture. In the "baseline" architecture referred to in these graphs, WebSphere is used with DB2. The same architecture was then used with enJin added to the middle tier between WebSphere and DB2.
The first graph shows response times for simple transactions. These are new trades in the stock trading benchmark. At 10 test-harness threads, the baseline configuration response time was 900 milliseconds compared to 160 milliseconds for the configurations with enJin.
The second graph shows response times for more complex transactions. These are new portfolio transactions in the stock trading benchmark. At 10 test-harness threads, the baseline configuration response time was 3,730 milliseconds compared to 600 milliseconds for the configurations with enJin.
So, it is pretty easy to make the case that with this accelerator in the middle tier, you can either have higher capacity with the same hardware or you can use less expensive hardware. The white paper gives an example of such a calculation.
Note: Similar results can be achieved using an XML database in the middle tier. An architecture along with performance charts similar to the two shown above can be found in Chapter 5, Solve Performance Problems with FastSOA Patterns, in FastSOA by Frank Cohen.
? The information on this benchmark is used with permission from Versant Corporation. The full benchmark paper is available here.
More on the general topic: Architectures Using Databases in the Middle Tier
- J2EE Enterprise Architecture with In-Line Updates
- J2EE Enterprise Architecture with Staged Updates
- J2EE Website Architecture