### Table 1. Exhaustive generation.

"... In PAGE 8: ... The average and maximal sizes are shown in Table 1. Table1 and Figure 6 lead to some observations: 1. A very small number of patterns produce BM automata with maximal size.... In PAGE 10: ... However, a random generation is appropriate for a good approximation of the average sizes. We applied the least square method to the 22 average sizes of Table1 and the 10 estimated average sizes of Table 2 (obtained by a uniform random generation of max(1000; p2m) patterns, for each length m, 23 m 32). We then obtained the best approximated curve (see Figure 7) ave size(m) 0:12m3 + O(m2): Note that this average is about twice the cubic lower bound already presented in Proposition 2.... In PAGE 10: ...ig. 7. Maximal and average sizes. We also considered the case where the text had more than two letters but the pattern was restricted to two letters. In this case the results are almost the same as in Table1 with the exception of m = 7 (maximal size is 44 for bbabbbb), m = 8 (maximal size is 62 for bbabbbbb) and m = 10 (the maximum is also reached by bbabbbbbbb). The di erence is due to the fact that when there is a symbol di erent from a and b in the text, a shift has to jump over that symbol, generating states which remember a pre x of the pattern that is also a su x.... ..."

### Table 4.3 Comparison of Iterative Deepening and direct search

"... In PAGE 12: ...able 4.15 Search depths for different stages ...................................................................75 Table4 G17G20G25G3G51G68G85G68G80G72G87G72G85G86G3G68G15G3G69G3G68G81G71G3 G3G72G86G87G76G80G68G87G72G71G3G69G92G3G79G76G81G72G68G85G3G85G72G74G85G72G86G86G76G82G81 .... ..."

### Table 3. Results on large structures with very complex structural integrity constraints. All times are in milliseconds.

"... In PAGE 16: ... We further test the scalability of Shekoosh when generating structures with very complex constraints on the structure (complex structures truly determine the efficiency of the repair algorithm and thus the generation approach). Table3 shows the results for generating the structures in category (3). Note that as the structure size increases, both Egor and Dicos scale linearly , yet the repair algorithm grows faster.... ..."

### Table 1. Worst efficiency bounds for scheduling policies.

2000

"... In PAGE 8: ... The behavior of the policies was very similar for all the number of workers, but it was strongly affected by the variation of the execution times of the tasks in different iterations, by the workload and by having significant differences among the execution times of the 20% largest tasks. Table1 shows the efficiency bounds obtained for the previously described scheduling policies, always relative to LPTF policy. The first column contains the upper bound that is never surpassed in 95% of cases.... ..."

Cited by 5

### Table 1. Worst efficiency bounds for scheduling policies.

"... In PAGE 8: ... The behavior of the policies was very similar for all the number of workers, but it was strongly affected by the variation of the execution times of the tasks in different iterations, by the workload and by having significant differences among the execution times of the 20% largest tasks. Table1 shows the efficiency bounds obtained for the previously described scheduling policies, always relative to LPTF policy. The first column contains the upper bound that is never surpassed in 95% of cases.... ..."

### Table 2. Symbols and function names of data extracted from logs for iterative deepening

2002

"... In PAGE 5: ... In the definitions of CAB4C9BN D2B5 and C5B4C9BN D2B5, note that a single Response message can hold more than one result. We also estimated several values, listed in the second half of Table2 . These values could not be directly observed, but they are nevertheless carefully calculated from observed quantities (see [20] for details).... In PAGE 6: ... For example, a Query message contains a Gnutella header, a query string, and a field of 2 bytes for options. Headers in Gnutella are 22 bytes, TCP/IP and Ethernet headers are 58 bytes, and the query string is C4B4C9B5 bytes ( Table2 ). Total message size is therefore 82 + C4B4C9B5 bytes.... ..."

Cited by 97

### TABLE IV Performance and computation time of the agents using random, independent, and iterative deepening algorithm in the 6 vs. 6 setting

1999

### Table 2. Symbols and function names of data extracted from logs for iterative deepening

"... In PAGE 5: ... In the definitions of CAB4C9BN D2B5 and C5B4C9BN D2B5, note that a single Response message can hold more than one result. We also estimated several values, listed in the second half of Table2 . These values could not be directly observed, but they are nevertheless carefully calculated from observed quantities (see [20] for details).... In PAGE 6: ... For example, a Query message contains a Gnutella header, a query string, and a field of 2 bytes for options. Headers in Gnutella are 22 bytes, TCP/IP and Ethernet headers are 58 bytes, and the query string is C4B4C9B5 bytes ( Table2 ). Total message size is therefore 82 + C4B4C9B5 bytes.... ..."

### TABLE IV Performance and computation time of the agents using random, independent, and iterative deepening algorithm in the 6 vs. 6 setting

1999

### Table 6: Randomly generated networks with no structural constraints: Bound=16.

1998

"... In PAGE 37: ... A hypothesis consistent with the data is that, in general, the more structured and more con- strained a network is, the smaller the minimal, equivalent, dispatchable network is. The networks at Table 2 and Table6 both have relatively loose constraints (the bound of the edges at table 2 were chosen randomly and uniformly from the interval [0, 10000] while the ones at Table 6 from the interval [0,15]). However, the average ratio of output edges per nodes is 2.... In PAGE 37: ... A hypothesis consistent with the data is that, in general, the more structured and more con- strained a network is, the smaller the minimal, equivalent, dispatchable network is. The networks at Table 2 and Table 6 both have relatively loose constraints (the bound of the edges at table 2 were chosen randomly and uniformly from the interval [0, 10000] while the ones at Table6 from the interval [0,15]). However, the average ratio of output edges per nodes is 2.... ..."