Memory Consumption Tests

    Although the ServiceBinder eases considerably OSGi programming, its use comes with a cost that reflects itself as an increase in memorry consumption. To estimate memory consumption, a series of tests were conducted based on the techniques discussed in [Wilson2000]. These tests are based on a technique which involves creating a considerable amount of objects and then dividing the memory increase by the number of objects. Every test was conducted ten times with the garbage collector disabled. The tests were realized on a 700Mhz Linux PC with 512Mb of memory, and on top of the OSCAR platform running in a Blackdown JVM. Three tests were conducted:

• Test No. 1 estimates memory consumption for a component that registers a single service. An equivalent OSGi service provider was created to do the comparison. In this case, the memory overhead is 381 bytes, which represents a 244% increase over the standard case.

• Test No. 2 estimates memory consumption for a component that registers a single service and requires a service (0..n, dynamic). In this case the memory overhead is 480 bytes, which represents a 55% increase over an equivalent entity programmed directly in OSGi.

• Test No. 3 estimates memory consumption for the previous case when additionnal services are required. In this case, the memory overhead is 264 bytes, which represents a 34% increase over the standard case.

    The results from these tests are shown on figure 1. The memory overhead for test no. 1 may seem excessive, but it can be explained by the fact that to register one service, the component descriptor must be reified (put into memory) and an instance manager must be created. Tests no. 2 and 3 show that memory overhead is not excessive when services are required. We think that the overhead is acceptable when taking into account the benefits that the ServiceBinder brings with respect to simplifying development.