The performance requirements where met in production. therefore closing.

PS: By optimizing the aggregation calculation (cf. NMS-13100) the computational load is reduced.

In addition to the existing 4 aggregations

• 1: exporter/interface

• 3: exporter/interface/(app|conv|host)

4 more aggregations are now stored:

• 1: exporter/interface/tos

• 3: exporter/interface/tos/(app|conv|host)

The existing aggregations are kept for backwards compatibility.

The size of the new aggregations differ from the sizes of the old aggregations by a certain factor. The upper bound for that factor is 192 (=64 * 3) because in theory there are 6 bits that distinguish DSCPs and 3 different kinds of ECNs (encoded by 2 bits). In practice however, the factor should be considerably smaller because only 18 different DSCP codes are expected. The storage cost depends on the actual existence of different DSCP codes.

If ECN information should not be used to filter/aggregate then the current implementation can be configured to ignore the ECN information. This would reduce the upper bound of the increased storage cost factor from 192 to 64.

The increase in storage cost also corresponds to increased writes to elastic.

The computational cost (in Nephron) increases, too. For the current implementation I expect the computational cost to double. This guess is based on the fact that the number of aggregations doubles. Future work may reduce the computational cost because aggregations can be build on top of each another.