"... In PAGE 13: ... We first compare the different random, DHT, and clairvoyant data placement strategies. Table5 shows for all the placement strategies the total number of repairs triggered, average band- width per node, and percentage of improvementover Random. Additionally, Table 5 the average and standard deviation of the number of replicas per node.... In PAGE 13: ... Table 5 shows for all the placement strategies the total number of repairs triggered, average band- width per node, and percentage of improvementover Random. Additionally, Table5 the average and standard deviation of the number of replicas per node. The storage system parameters were replication redundancy scheme m = 1, minimum thresh- old number of replicas th = 5, and a heartbeat timeout to = 1hr.... In PAGE 13: ... The size of the blacklist for the RandomBlackList and RandomSiteBlacklist placement strategies was the top 35 nodes with the longest total downtimes. Table5 shows how the data placement strategies differ in cost for the estimated optimal number of extra replicas e = 6; thus, the total number of replicas per object was n = 11. Table 5 shows that more sophisticated placement strate- gies exhibited noticeable increase in performance; that is, fewer repairs triggered compared to Random.... In PAGE 13: ... Table 5 shows how the data placement strategies differ in cost for the estimated optimal number of extra replicas e = 6; thus, the total number of replicas per object was n = 11. Table5 shows that more sophisticated placement strate- gies exhibited noticeable increase in performance; that is, fewer repairs triggered compared to Random. For ex- ample, the RandomSiteBlacklist placement showed a 4.... ..."

"... In PAGE 13: ... We first compare the different random, DHT, and clairvoyant data placement strategies. Table5 shows for all the placement strategies the total number of repairs triggered, average band- width per node, and percentage of improvementover Random. Additionally, Table 5 the average and standard deviation of the number of replicas per node.... In PAGE 13: ... Table 5 shows for all the placement strategies the total number of repairs triggered, average band- width per node, and percentage of improvementover Random. Additionally, Table5 the average and standard deviation of the number of replicas per node. The storage system parameters were replication redundancy scheme m = 1, minimum thresh- old number of replicas th = 5, and a heartbeat timeout to = 1hr.... In PAGE 13: ... The size of the blacklist for the RandomBlackList and RandomSiteBlacklist placement strategies was the top 35 nodes with the longest total downtimes. Table5 shows how the data placement strategies differ in cost for the estimated optimal number of extra replicas e = 6; thus, the total number of replicas per object was n = 11. Table 5 shows that more sophisticated placement strate- gies exhibited noticeable increase in performance; that is, fewer repairs triggered compared to Random.... In PAGE 13: ... Table 5 shows how the data placement strategies differ in cost for the estimated optimal number of extra replicas e = 6; thus, the total number of replicas per object was n = 11. Table5 shows that more sophisticated placement strate- gies exhibited noticeable increase in performance; that is, fewer repairs triggered compared to Random. For ex- ample, the RandomSiteBlacklist placement showed a 4.... ..."

"... In PAGE 26: ...3. Summary In Table3 we have summarized the probability of data not being availabile and the impact of a processor... ..."

"... In PAGE 26: ...3. Summary In Table3 we have summarized the probability of data not being availabile and the impact of a processor... ..."

"... In PAGE 29: ... Even though the Mariposa optimizer will pick the same join order as the tra- ditional optimizer, the broker can change its query processing strategy because it acquires bids for the two joins among the 3 sites. Examination of Table7 reveals the performance improvements resulting from dynamic movement of objects. It shows the elapsed time, location of data and revenue generated at each site by running the 3-way join query described in Section 6.... In PAGE 29: ... The experimental results vary considerably because of the wide variance in Internet network latency. Table7 shows a set of results which best illustrate the beneficial effects of the economic model. 7.... ..."

"... In PAGE 29: ... Even though the Mariposa optimizer will pick the same join order as the tra- ditional optimizer, the broker can change its query processing strategy because it acquires bids for the two joins among the 3 sites. Examination of Table7 reveals the performance improvements resulting from dynamic movement of objects. It shows the elapsed time, location of data and revenue generated at each site by running the 3-way join query described in Section 6.... In PAGE 29: ... The experimental results vary considerably because of the wide variance in Internet network latency. Table7 shows a set of results which best illustrate the beneficial effects of the economic model. 7.... ..."

"... In PAGE 14: ... Even though the Mariposa optimizer will pick the same join order as the traditional optimizer, the broker can change its query processing strategy because it acquires bids for the two joins among the three sites. Examination of Table7 reveals the performance improvements resulting from dynamic movement of objects. It shows the elapsed time, location of data and revenue generated at each site by running the three-way join query described in Sect.... In PAGE 14: ... The experimental results vary considerably because of the wide variance in Internet network latency. Table7 shows a set of results which best illustrate the beneficial effects of the economic model. 7 Related work Currently, there are only a few systems documented in the literature that incorporate microeconomic approaches to re- source sharing problems.... ..."

"... In PAGE 29: ... Even though the Mariposa optimizer will pick the same join order as the tra- ditional optimizer,the broker can change its query processing strategy because it acquires bids for the two joins among the 3 sites. Examination of Table7 reveals the performance improvements resulting from dynamic movement of objects. It shows the elapsed time, location of data and revenue generated at each site by running the 3-way join query described in Section 6.... In PAGE 29: ... The experimental results vary considerably because of the wide variance in Internet network latency. Table7 shows a set of results which best illustrate the beneficial effects of the economic model. 7.... ..."

"... In PAGE 6: ... Further, we do not assume any scope for parallelism in the execution, though there could be in reality (this only makes us conservative in our estimate of the benefits of work partitioning), and w4 is set to zero in our considerations. Table1 shows the work partitioning strategies that are actually studied in this paper. The taxonomy is based on... In PAGE 14: ...1 Adequate Memory at Client We first consider the scenario where the client has adequate memory to hold all of the dataset and index if needed. As is pointed out in Table1 , even in this scenario, we consider some situations where the actual data objects are not necessarily present and need to be shipped from the server after a refinement. 6.... ..."