Skip to main content

Table 1 Comparison of our three proposed architectures. k corresponds to the number of different sensor-group-pairs, e is the emit rate, w is the window size, and g is the grace period

From: Scalable and Reliable Multi-dimensional Sensor Data Aggregation in Data Streaming Architectures

Architecture Basic Tumbling window–based Hopping window–based
Support for out-of-order records
Unknown record frequency
Updates triggered per record 1 1 \(\frac {w}{e}\)
Memory usage \(\mathcal {O}(k)\) \(\mathcal {O}(k \frac {g + e}{e})\) \(\mathcal {O}(k \frac {g + w}{e})\)
  1. Note that for the tumbling window–based architecture, e equals w. Whereas our basic architecture does not support out-of-order records and the tumbling window–based architecture requires the measurement frequency of sensors to be known beforehand, the hopping window–based architecture is able to handle both. However, in contrast to the former two, it generates significantly more intermediate records and thus uses more memory in the last value table.