Abstract: Precise quantile determination in a distributed environment is provided where data is distributed among worker nodes, one of which functions as a coordinator node. Each worker node has local data for precise quantile determination. The coordinator node derives a maximum relative error value ? which guarantees an exact result for every quantile can be determined in two passes with optimal memory consumption on this node. The worker nodes perform a distributed computation of quantile summaries of datapoints, that specify lower and upper bounds of quantile ranks for selected datapoints. The coordinator node merges the quantile summaries and determines, for each quantile rank, lower and upper bounds for values between which the quantile must occur. The worker nodes filter out, in a second pass, those datapoints that are not within the bounds and stream the qualifying data to the coordinator node. The coordinator node computes precise quantiles from the qualifying data.

Abstract: A system for precise quantile determination derives a maximum relative error value ? that guarantees that an exact result for every quantile can be determined in no more than two passes. The system computes a quantile summary of the datapoints, that specifies lower and upper bounds of the quantile ranks for selected datapoints, and determines for each quantile rank, the lower and upper bounds for values between which the quantile must occur. The system filters out in a second pass those datapoints that are not within the bounds to provide subsets; and computes from the quantile ranks and the subsets, the precise quantiles.

Abstract: Precise quantile determination in a distributed environment is provided where data is distributed among worker nodes, one of which functions as a coordinator node. Each worker node has local data for precise quantile determination. The coordinator node derives a maximum relative error value ? which guarantees an exact result for every quantile can be determined in two passes with optimal memory consumption on this node. The worker nodes perform a distributed computation of quantile summaries of datapoints, that specify lower and upper bounds of quantile ranks for selected datapoints. The coordinator node merges the quantile summaries and determines, for each quantile rank, lower and upper bounds for values between which the quantile must occur. The worker nodes filter out, in a second pass, those datapoints that are not within the bounds and stream the qualifying data to the coordinator node. The coordinator node computes precise quantiles from the qualifying data.

Abstract: Precise quantile determination in a distributed environment is provided where data is distributed among worker nodes, one of which functions as a coordinator node. Each worker node has local data for precise quantile determination. The coordinator node derives a maximum relative error value ? which guarantees an exact result for every quantile can be determined in two passes with optimal memory consumption on this node. The worker nodes perform a distributed computation of quantile summaries of datapoints, that specify lower and upper bounds of quantile ranks for selected datapoints. The coordinator node merges the quantile summaries and determines, for each quantile rank, lower and upper bounds for values between which the quantile must occur. The worker nodes filter out, in a second pass, those datapoints that are not within the bounds and stream the qualifying data to the coordinator node. The coordinator node computes precise quantiles from the qualifying data.

Abstract: A system for precise quantile determination derives a maximum relative error value ? that guarantees that an exact result for every quantile can be determined in no more than two passes. The system computes a quantile summary of the datapoints, that specifies lower and upper bounds of the quantile ranks for selected datapoints, and determines for each quantile rank, the lower and upper bounds for values between which the quantile must occur. The system filters out in a second pass those datapoints that are not within the bounds to provide subsets; and computes from the quantile ranks and the subsets, the precise quantiles.