"... In PAGE 14: ... 4.4 Eliminating philosophers In Table4 we show how it is possible to extend the speci cation of dining philosophers by introducing also the possibility of eliminating philosophers ac- tually around the table. Also the process of elimination can start only when a philosopher is thinking.... ..."

"... In PAGE 10: ... The spawning performance test methodol- ogy is shown in Figure 5: a process repeatedly spawns itself across two nodes to work out the cost of an individual spawn operation. The spawn operation has been implemented in seven di erent ways (refer forward to Table4 for a list). The di erent types of spawning should be self explanatory from their names except: RUNON is a MOSIX command, which can be used to start a remote process on a particular machine, just like rsh, or ssh.... In PAGE 14: ...2 Dynamic creation measurements This section benchmarks the time to \spawn quot; a new process using a variety of di erent mechanisms. The timings on the Linux cluster are given in Table4 . Additionally, timings on a 1.... ..."

"... In PAGE 25: ... Actual dynamical implementation of the correction mechanism is a focus of current research. The Input Map for the Dynamical Automaton we used to model transitive sentences with relative clause modi ers is shown in Table3 . The automaton uses 9 partition states and moves around on a 3-dimensional fractal.... In PAGE 30: ...De nition Start (1/2, 1/2, 1/2) Comp1 (0, 0, 0) + opencube Comp2 (0, 1/2, 0) + opencube V1 (1/2, 0, 0) + opencube V2 (1/2, 1/2, 0) + opencube NObj1 (0, 0, 1/2) + opencube NObj2 (0, 1/2, 1/2) + opencube NSubj1 (1/2, 0, 1/2) + opencube NSubj2 (1/2, 1/2, 1/2) + opencube Note: opencube is the set f(x; y; z) : 0 lt; x lt; 1=2; 0 lt; y lt; 1=2; 0 lt; z lt; 1=2g. Note: Compartment A as labelled in Table3 is the union of the compartments A1 and A2 shown above for A 2 fComp; V; NObj; NSubjg. Table 4: Dynamical Automaton 2: compartment de nitions.... ..."

"... In PAGE 5: ... Table3 . State table for a fork structure As a summary, the flexible DFA-based implementation enables the dynamic reconfiguration of the query graph and switching optimization goals at run time.... ..."

"... In PAGE 16: ... The fault parameters are: when - following the last normal event, what - generate an extra message, how- provide an extra message masked as ProgNOK, where - in PCO2. The abstract fault case for this scenario is shown in Table5 in TTCN notation. Table 5 Dynamic Behavior Description Fault Case Test Case Dynamic Behavior Test Case Name : state execution report with success and failure in the progress... ..."

"... In PAGE 12: ... The same job stream was run on a Model 148 configured like the Model 145. The results, summarized in Table4 , provide insight into the effects of VM assist and ECPS/VM during CMS execution. IV.... ..."

"... In PAGE 3: ... A philosopher object tries to perform a state transition whenever its \tick quot; method is invoked. Fork objects are created from the class in Table1 . Philosopher objects are created from the classes in Table 2: cphil lr for philosophers that pick up their left fork before their right, and cphil rl for the other type of philosopher.... ..."

"... In PAGE 5: ... We now examine the effectiveness of compiler algorithms in eliminating synchronization. Table3 displays the number of par- allel loops (doalls) and barriers executed dynamically by each program at run time, and the percentage eliminated by different levels of optimization. The first three columns for doalls indi- cate the numberof parallel loops executedin the original program, the percentageeliminated by merging adjacentdoallsinto thesame parallel SPMD region, and the percentage eliminated by aggres- sivelyexpandingSPMDregionstoincludetheoutertime-steploop.... In PAGE 5: ...eighbor synchronization. Optimizations are cumulative. Dashes are used to mark programs when no optimizations have been ap- plied. We see from Table3 that the compiler is effective at reducing the number of parallel tasks and barriers actually executed by the application at run time. We find the compiler is quite successful in discovering adjacent parallel loops which may be merged into a singleparallel region, eliminating onaverage88%ofparallel invo- cations and 22% of barriers executed.... ..."

"... In PAGE 6: ... We define a set of tractable modifications to registers which we can easily reverse the effect of and calculate earlier register values based on a final end-of-basic block register file snapshot. Table6 lists the four integer instructions we support as well as the symbolic execution that updates register values, Any other instructions that assign to an integer register are defined as in- tractable. If and only if a register used for addressing is modified by an intractable instruction in the middle of the basic block, we insert a snippet to store its value before modification.... ..."

"... In PAGE 5: ... Due to space constraints, we only report on the ten largest results in each experiment. We took the ten programs that took longest to compile using the standard SSA-to-CFG con- version for Table 2 and Table 3 and the ten programs with the most dynamic copies for Table4 and Table 5. However, we think that these give a reasonable insight into the behavior of the algorithms.... In PAGE 6: ... 4.3 Efficacy Measurements In Table4 , we show the number of copy operations that were executed. Our algorithm produces code that executes about 1% fewer copies, on average, than the interference-graph coalescer.... ..."