In high performance scalable systems the optimal utilization
of memory resources is essential for both performance and
energy efficiency. In this paper, we propose to the best of
our knowledge for the first time a new hardware-oblivious,
software-based, online working set change detection algo-
rithm for each process in a time-shared system. This novel al-
gorithm is intended to be used as crucial trigger for essential
decision making procedures, such as launching reclamation
or compaction algorithms, collecting memory-related statis-
tical data, and many more. The proposed solution perceives
the occurrence of concavity in a program’s lifetime function
as an outlier and heuristically attempts to classify such from
the normal observations. Our novel online algorithm detects
the working set change with incomplete knowledge of a
task’s memory access behavior. The proposed online algo-
rithm is implemented and evaluated in the Linux kernel, and
executes with O (1) time and space complexity. Evaluations
using real-world data-intensive applications show that the
algorithm detects the working set change with an average
accuracy of 90% and F0.5-Score of 77%, adding insignificant
computation overhead averaging around 1.08E-06%.