"... In PAGE 18: ... 1991; Peterson 1986; Tatar 1988; 19901. (Additional references are listed in Table3 .) Using terms such as com- puter-supported cooperative work and groupware, these systems perform functions such as helping people collabo- rate on writing the same document, man- aging projects, keeping track of tasks, and finding, sorting, and prioritizing electronic messages.... In PAGE 20: ...3.7 A Taxonomy of Cooperative-Work Tools As shown in Table3 , the framework we have suggested for coordination provides a natural way of classifying existing cooperative-work systems according to the coordination processes they support. Some of these systems primarily empha- size a single coordination-related process.... ..."

"... In PAGE 32: ...0 integrated with eRoom Real Time Services. Table1 _ Enhancements to CASE tools Our study concluded that no single tool came close to covering the needs of GST. Also, no combination of tools could provide an ideal environment.... In PAGE 38: ....1. Influencing factors We base our analysis on a taxonomy of 16 influencing factors of distributed and interdisciplinary development projects according to [3]. Table1 shows these factors with their parameter values. It combines ... In PAGE 39: ...[4][5][6], as well as the factors resulting from the interaction of different domains presented in [7]. Table1 . Influencing factors 2.... In PAGE 39: ...2. Empirical study In the empirical study 18 standard IT-tools which support global development projects have been rated by experts for their support of cooperation projects with respect to the influencing factors of Table1 . The survey was carried out by interviewing experts in distributed projects, interdisciplinary projects, cooperation projects, and Information Technology (IT).... In PAGE 39: ... Our goal is to determine with the help of the expert data which factors have more impact on the selection of each tool. The input factors are shown in Table1 . Their value sets are derived from the results of the survey in [8].... In PAGE 46: ... Role-Oriented Notification System In order to fully address the REQM needs GSD project participants need a generic mechanism to ensure timely notifications about changes to requirements or other artifacts in addition to the common data exchange for- mat mentioned above [13][14]. Multiple requirements- related events may happen concurrently at several sites across a GSD project (see overview in Table1 ), which would seem relevant for project team members at other sites. For example, when requirement 4711 in the re- quirements management tool changes (event in the requirements management tool), the developer who works on this requirement can be notified, e.... In PAGE 47: ...ule set; e.g., which rules should be active. Events in the REQM tool new requirement is inserted Existing requirement is changed New user is added User privileges are changed Events in the IDE Trace is created between source code element and re- quirement Events in the Configuration Management tool New change request was submitted State of an artifact changed Check-in of an artifact Events in the test management tool Bug report is inserted Test cases for a particular requirement are reported as successfully tested. Table1 . Examples for events in GSD tool support.... In PAGE 52: ... The timeframe for the MaPIT research project is three years, spanning from January 2007 to December 2009. The preliminary schedule for the proposed research is shown on Table1 . Schedule Date Activity 01/2007 Initiation of the research project 03/2007 - 09/2007 First interview round 05/2007 - 11/2007 Analysis of first interview round 01/2008 - 04/2008 Second interview round 03/2008 First survey 06/2008 Stage 1 research data col- lected 09/2008 - 11/2009 Consecutive interview rounds 03/2009 Second survey 12/2009 Stage 2 research data col- lected Table 1.... In PAGE 52: ... The preliminary schedule for the proposed research is shown on Table 1. Schedule Date Activity 01/2007 Initiation of the research project 03/2007 - 09/2007 First interview round 05/2007 - 11/2007 Analysis of first interview round 01/2008 - 04/2008 Second interview round 03/2008 First survey 06/2008 Stage 1 research data col- lected 09/2008 - 11/2009 Consecutive interview rounds 03/2009 Second survey 12/2009 Stage 2 research data col- lected Table1 . The proposed research will be made as a part of MaPIT project, and thus its schedule follows the schedule of the research project.... In PAGE 58: ...2. Use of the coordination tools Table1 shows the number of proposals (requests of turn, compilation proposals, and execution proposals) and answers made with the coordination tools. There are two surprising results.... ..."

"... In PAGE 2: ...In this paper, we continue the SAF work by developing an SAF protocol that, while reliable, is complexity-wise as e cient as the unreliable SAF protocol (see Table1 ). Our major contribution is the design of a novel and e cient reli- ability method for use by hardware-based ooding.... ..."

"... In PAGE 47: ... For tree topologies, the main variables involve the number of branches (Nb) at each node of the tree, and which branch is sent to rst. These classes are explained in detail below, and Table7 provides a quick summation of some of the more important properties. This Table speci es the number of steps until the algorithm completes (STEPS), the number of messages sent during step i (SENDS, S = i), the number of processors who are nished with the routine after step i is complete (PROCS DONE, S = i), the time the source processor spends in the algorithm (SRC TIME), and nally the maximum time spent by any processor in the operation (MAX TIME).... In PAGE 47: ... This Table speci es the number of steps until the algorithm completes (STEPS), the number of messages sent during step i (SENDS, S = i), the number of processors who are nished with the routine after step i is complete (PROCS DONE, S = i), the time the source processor spends in the algorithm (SRC TIME), and nally the maximum time spent by any processor in the operation (MAX TIME). The analyses shown in Table7 have been simpli ed by assuming that Nr is an even multiple of Np, and Nb = 1, with Np an integer multiple of 2. The speci c topology section should be examined for full details.... In PAGE 50: ... With this algorithm, Np does not have to be an integer power of Nb. The timing analysis for this algorithm is relatively complex, so we do not reproduce it here (analysis for the most common use, Nb = 1 is shown in Table7 ). See [15] for full details.... ..."

"... In PAGE 52: ... For tree topologies, the main variables involve the number of branches (Nb) at each node of the tree, and which branch is sent to rst. These classes are explained in detail below, and Table7 provides a quick summation of some of the more important properties. This Table speci es the number of steps until the... In PAGE 53: ...the number of processors who are nished with the routine after step i is complete (PROCS DONE, S = i), the time the source processor spends in the algorithm (SRC TIME), and nally the maximum time spent by any processor in the operation (MAX TIME). The analyses shown in Table7 have been simpli ed by assuming that Nr is an even multiple of Np, and Nb = 1, with Np an integer multiple of 2. The speci c topology section should be... In PAGE 56: ... With this algorithm, Np does not have to be an integer power of Nb. The timing analysis for this algorithm is relatively complex, so we do not reproduce it here (analysis for the most common use, Nb = 1 is shown in Table7 ). See [16] for full details.... ..."

"... In PAGE 3: ... All experiments in this work were conducted in single user mode. Table2 contains the measured timings with varying number of processors and problem sizes when using CMTM to broadcast vectors and also to re- duce by summation. As expected, MMPI Reduce is always more time consuming than... ..."

"... In PAGE 14: ... Result 1 Subjects apos; actions are in uenced to some extent, but not completely, by strategic consid- erations. The third column of Table2 shows the aggregated (all subjects, all rounds) relative frequencies of cooperation in each cell of the experiment. Notice that di erences in the level of cooperation are much more stark between games than between information conditions; cooperation is generally highest in the SH cells and lowest in the PD cells.... In PAGE 15: ... Result 2 The addition of either observation or communication increases the frequency of coopera- tion relative to the control sessions, though the relative e cacy of observation and communication depends on the game. Supporting evidence for this second result is provided in Table2 . Consider rst the PD cells.... In PAGE 16: ...19 In the SH cells, there are no signi cant di erences between treatments. This lack of signi cance may seem surprising, since Table2 suggests that adding either observation or cheap talk increases the overall frequency of cooperation substantially. The explanation for this discrepancy is that there is a lot of variance in the control sessions; in two of the three control sessions, the frequency of cooperation is close to 50%, while in the other one, it is much higher (81%) and comparable to that in the non{control cells.... In PAGE 17: ... Result 3 In the games with multiple equilibria, cheap talk aids successful coordination on a pure{ strategy Nash equilibrium. According to Table2 (fourth column), coordination in the SH game is most frequent in the cheap talk cell; it is signi cantly more likely than in the control at the 5% level (robust rank{order test), and it is signi cantly more likely than in the observation treatment at the 10% level. In the... In PAGE 19: ... A nal way in which the information treatment a ects aggregate behavior is its e ect on joint payo e ciency. (See Table2 , fth column.)... In PAGE 31: ... quot; According to any of the criteria discussed in the previous section, in Stag Hunt, words are relatively more useful than actions, while in Chicken and Prisoner apos;s Dilemma, actions are relatively more useful than words. It is worth noting that there is a correspondence between the usefulness of a set of signals (according to any of these criteria) and its ability to e ect high{payo outcomes; in all three games, the more useful set of signals is also the one that leads to higher payo s (recall the last column of Table2 ), though the di erence in payo s may not be signi cant. Indeed, it should not be surprising that both sets of signals lead to payo s that are higher than when no signals are available, since both actions and words are at least somewhat informative in 25A remark about other types of non{signals is in order here.... ..."

"... In PAGE 52: ... For tree topologies, the main variables involve the number of branches (Nb) at each node of the tree, and which branch is sent to rst. These classes are explained in detail below, and Table7 provides a quick summation of some of the more important properties. This Table speci es the number of steps until the... In PAGE 53: ...the number of processors who are nished with the routine after step i is complete (PROCS DONE, S = i), the time the source processor spends in the algorithm (SRC TIME), and nally the maximum time spent by any processor in the operation (MAX TIME). The analyses shown in Table7 have been simpli ed by assuming that Nr is an even multiple of Np, and Nb = 1, with Np an integer multiple of 2. The speci c topology section should be... In PAGE 56: ... With this algorithm, Np does not have to be an integer power of Nb. The timing analysis for this algorithm is relatively complex, so we do not reproduce it here (analysis for the most common use, Nb = 1 is shown in Table7 ). See [16] for full details.... ..."

"... In PAGE 6: ... Thus, the function which determines the subclassification is: g b = preS ((); b) Despite the fact that this function is not binary, associa- tivity again plays a key ro le in the subclassification of the classes Sequential and Broadcast. Table3 lists different forms of function g with their corresponding subclasses of Sequential and Broadcast. 4.... In PAGE 8: ... Remember that n, the depth of the recursion, is still determined by the list of local inputs as = [(); n times : : : ; ()]. For all subclasses of Sequential, shown in Table3 , ex- cept Identity, we use a processor network with log n pro- cessors in a row (repeat) which computes the result with time(n) = O(log n), costBrent(n) = O(log n) and pipe(n) = O(1). We assume n to be a power of 2.... In PAGE 9: ....3. Broadcast In this class we apply function g to the global input n times and return a list of intermediate results and the final re- sult whose implementation is known from the previous sec- tion. For all subclasses shown in Table3 , we use a tree-like processornetworkwith n processorswhich computes the re- sult with time(n) = O(log2 n), costBrent(n) = O(n) and pipe(n) = O(1). At each node, a function is applied which receives the input from its parent on the left side and pro- vides two outputs to its children on the right.... ..."