Check out the new USENIX Web site. next up previous
Next: Effects of Message and Up: Performance Evaluation Previous: Uniprocessor vs. SMP

Savings in Space

Thus far, we have shown that our transformation mechanism enhances the performance of algorithms in terms of the ACET and the WCET in both SMP and uniprocessor environments. Here, we present results to support our claim that the transformation mechanism not only reduces the time overheads but also the space overheads of the algorithms. This is shown in Figure 14. Again, we varied the percentage of fast readers in the reader set. As expected, the amount of improvement increases as the number of fast readers increases. Moreover, the synchronization-based algorithm requires the least space, since it only needs a single shared message buffer. The NBW protocol and lock-based IPC, therefore, represent the extreme cases for the tradeoff between space requirements and WCET. The non-blocking IPC mechanisms, especially with our transformation, provide a good compromise, balancing WCET and memory usage.

Interestingly, the percentage of space reduction for all three transformed algorithms is the same, as shown in Figure 15(c). This does make sense since the memory requirements of the three original algorithms are all proportional to the number of readers in the reader set. So, the memory used by the transformed algorithms decreases proportionally to the number of slow readers. The slight variations in Figure 15(c) are due to some of the control variables that do not scale with the number of reader tasks. Overall, we achieve a reduction in memory usage that ranges from 14 to 70%.


next up previous
Next: Effects of Message and Up: Performance Evaluation Previous: Uniprocessor vs. SMP
hai huang 2002-03-30