Fig. 1. Comparison between real and estimated execution times for 50 WT10g queries. Parameter values: ti = 1400 ms, ts = 0.03 ms, tr = 4.0208 μs, tc1 = 0.00013068, tc2 = 0.000096.

Fig. 2. Linear and quadratic fit to vocabulary occurrence data.

Fig. 3. Log–log plot of unique vocabulary terms vs. total terms, fit to a linear model.

Fig. 4. Comparison of the response times for the uniform and skewed query models with the analytical model.

Fig. 5. Distributed IR model.

Fig. 6. Throughput for the simulation of a distributed IR system with local index organisation.

Fig. 7. Throughput (queries per second) for different replicated IR systems versus the total number of query servers in the system.

Fig. 8. Network load for different replicated IR systems versus the number of query servers per replica.

Fig. 9. Throughput for several replicated IR systems, with a 10% of compression ratio. The stronger line represents a 10% compression and the thinner line is the baseline (without compression).

Fig. 10. Clustered IR systems vs. a replicated IR system. Configuration 1: 128 query servers.

Fig. 11. Clustered IR systems vs. a replicated IR system. Configuration 2: 1024 query servers.

Parameters for the document model

Parameters for the query model

Parameters for the distributed model

Estimated time (in min) to process 50 queries by the distributed IR system with 3 brokers

Throughput (queries per second) for different replicated IR systems

The “Query servers” column represents the number of servers per replica. Each column indicates the number of replications (R) and the number of brokers used (B).

Average throughput (T) and average send time per byte in seconds (S), using different compression ratios, in a distributed IR system (one replica)

Distribution of queries across general topic categories, and configurations for the clustered systems simulated

Average throughput (queries/s) for the basic sub-collection (8.5 million documents), in a distributed IR system