"... In PAGE 4: ... 5.2 Declarations of function pointers Table2 deals with the questions regarding declaration of function pointers and the scopes within which they are declared. In Program, we list the name of the program and its lines of code.... ..."

"... In PAGE 8: ...2), AIRPORT is the concurrency controller for an Airport Ground Traffic Control simulation program mentioned in Section 3, BB is a bounded buffer, BB-MUTEX is a bounded buffer with a mutex lock, BB-RW is a bounded buffer with a reader-writer lock, and BARRIER is a controller for bar- rier synchronization. Table2 shows the performance results for the verification of these concurrency controllers. The columns labeled T1 and M1 denote the time and the mem- ory usage for the behavior verification by the Action Lan- guage Verifier.... In PAGE 8: ... Hence, we do not need to consider all possible thread interleavings during interface verification. In Table2 , the last two columns (T2 and M2) show the time and memory usage of JPF during the interface verifi- cation. The memory usage is low since we use stubs with fi- nite reachable state spaces and we do not have to consider all possible thread interleavings.... In PAGE 8: ... The memory usage is low since we use stubs with fi- nite reachable state spaces and we do not have to consider all possible thread interleavings. For the examples reported in Table2 we verified the thread implementations by run- ning a single instance of each thread class in isolation. JPF successfully verifies the problem instances with stubs with- out exhausting memory.... ..."

"... In PAGE 42: ... Only three of these nodes are used as benchmarks, namely a network TMOS server, a network coordinator, and the alarm server. Some additional information about the benchmarks is contained in Table1 . De- tailed statistics about message sizes can be found in [46].... In PAGE 43: ...PERFORMANCE EVALUATION Benchmark Processes Messages ring 100 100,000 life 100 800,396 eddie 2 2,121 BEAM compiler 6 2,481 NETSim TMOS 4,066 58,853 NETSim coordinator 591 202,730 NETSim alarm server 12,353 288,675 procs 100x100 100 6,262 procs 1000x100 1,000 512,512 procs 100x1000 100 6,262 procs 1000x1000 1,000 512,512 Table1 : Number of processes and messages. a consistent picture, we decided to also use this machine for all other benchmarks too.... In PAGE 45: ... This is to some ex- tent expected, since NETSim is a commercial product developed over many years using a private heap-based Erlang/OTP system and tuned in order to avoid garbage collection and reduce send times. For example, from the number of processes in Table1 and the maximum total heap sizes which these programs allocate (data shown in Table 2), it is clear that in the NETSim programs either the majority of the processes do not trigger garbage collection in the private heap system as their heaps are small, or processes are used as a means to get no-cost heap reclamation. As a result, the possible gain from a different memory architecture cannot be big.... In PAGE 45: ... Indeed, as observed in the case of NETSim alarm server, the large root set (cf. Table1 ) can seriously increase the time spent in garbage collection and slow down execution of a program which has been tuned for a private heap architecture. We suspect that the general speedup for the mutator in the shared heap system is due to better cache locality: partly due to requiring fewer garbage collections by sharing data between processes and partly due to having heap data in cache when switching between processes.... In PAGE 69: ...02% nag - keep 1000x250 50,329,185 100% lt; 0.02% Table1 : Numbers of messages sent and (partially) copied in the hybrid system. 6.... In PAGE 69: ... 6.2 Effectiveness of the message analysis Table1 shows numbers of messages and words copied between the process-local heaps and the message area in the hybrid system, both when the message analysis is not used2 and when it is. In the life benchmark, we see that while there is hardly any reuse of message data, so that the plain hybrid system cannot avoid copying data from the local heaps to the shared area, when the analysis is used the amount of copying shrinks to zero.... In PAGE 71: ... The hybrid system with analysis is about 10% faster than the process-centric system, but we can see that although enabling the analysis removes the need for actual copying of message data (cf. Table1 ), we still have a small overhead for the runtime safety check performed at each send operation (this could in principle be removed), which is comparable to the total copying time in the process-centric system when messages are very small. We can also see how the slightly more complicated bookkeeping for sending messages is noticeable in the process-centric system, and how on the other hand the mutator time can be larger in the hybrid system.... In PAGE 93: ... Although this program takes a very small time to complete, we use it as a benchmark because it spawns a large number of simultaneously live processes (cf. Table1 ) and thus its root set is quite large. adhoc A framework for genetic algorithms.... ..."