"... In PAGE 19: ...s done by the thread library. But this makes them very flexible. It is easy to implement several scheduling algorithms in the thread library in order to customize user threads for special applications. Table1 summarizes the advantages and disadvantages of user and kernel... ..."

"... In PAGE 3: ... This contrasts with Linux Thread, which invokes the sched yield() system call regardless of the number of threads. Performance of thread control Table3 summarizes the average time taken for several important methods of thread control. We see that Green Thread shows better performance than Linux Thread.... In PAGE 3: ... Then, the Java VM creates a new Thread ob- ject and initializes the detailed internal representation of a Java thread. The creation time in Table3 refers to the time needed for these processes. Once created, the Java thread Table 3: Time taken for several important thread controls.... ..."

"... In PAGE 27: ... The machine we used is a Sun UltraSparc T2000 (a Niagara) with 8 4-thread cores. In the performance results shown in Table2 , we did not use more than 8 cores since in that case, and due to access to shared units, speedups are sublinear even for completely independent tasks. The fork-join annotators we chose to compare with are MEL (Muthukumar et al.... In PAGE 28: ... In order to make the annotation unconditional (as the rest of the annotators we are dealing with), we simply removed the conditional parallelism in the places where it was not being exploited. This is why it appears in Table2 under the name UMEL. All the benchmarks executed were parallelized automatically by CiaoPP, start- ing from their sequential code.... In PAGE 29: ... Rather, our main fo- cus of attention is in the comparison among the speedups obtained using di erent annotators. A rst examination of the experimental results in Table2 , and also in Figure 12 allows inferring that in no case is UUDG worse than any other annotator, and in no case is UOUDG worse than (U)MEL. They should therefore be the annotators of choice if available.... ..."

"... In PAGE 9: ... We also provide library methods to specify several com- mon communication channel topologies, such as one, two and three dimensional grids or a binary tree (spmd thread 1dgrid, spmd thread 2dgrid, spmd thread 3dgrid, spmd thread tree). The library methods for thread communication are given in Table2 . The spmd produce and spmd consume methods are parameterized by a keyword specifying which thread the caller is being synchronized with.... ..."