Scalable Processing of Spatial Alarms

Abstract

Spatial alarms extend the idea of time-based alarms to the spatial dimension. Just as time-based alarms are set to remind us of the arrival of a future reference time point, spatial alarms are set on a spatial location of interest which the subscribers of the alarm will travel to sometime in the future. Spatial alarm processing requires meeting two demanding objectives: high accuracy, which ensures zero or very low alarm misses, and high scalability, which requires highly efficient and optimal processing of spatial alarms. In this paper we present a motion-aware framework, facilitated by two systematic methods, for scalable processing of spatial alarms. First, we introduce the concept of safe period to minimize the number of unnecessary spatial alarm evaluations, increasing the throughput and scalability of the system. We show that our safe period-based alarm evaluation techniques can significantly reduce the server load for spatial alarm processing compared to the periodic evaluation approach, while preserving the accuracy and timeliness of spatial alarms. Second, we develop a suite of spatial alarm grouping techniques based on spatial locality of the alarms and motion behavior of the mobile users, which reduces the safe period computation cost for spatial alarm evaluation at the server side. We evaluate the scalability and accuracy of our approach using a road network simulator and show that the proposed motion-aware safe period-based approach to spatial alarm processing offers significant performance enhancements for the alarm processing server while maintaining high accuracy of spatial alarms, especially compared to the conventional periodic alarm evaluation approach.

Scaling Spatial Alarm Services on Road Networks

“…Then we compute the Euclidean distance between the current location of m (L m ) and each of the border points for all alarms and select the border point that is nearest to L m to compute the shortest distance from m to the alarm target region. According to this shortest distance and a velocity metric of m, such as the global maximum speed or the expected speed of m [3], we calculate the new hibernation time for m. Though this approach is easy to implement, its main weakness is the unnecessarily short hibernation time due to the use of Euclidean distance rather than road network distance. Consequently, mobile clients need to wake up frequently, consuming higher battery energy than necessary.…”

“…However, less frequent alarm checks may cause high alarm miss rate. The state of art techniques to spatial alarm processing is safe period [3] and safe region [2], [5] techniques, which use Euclidean distance between a mobile subscriber and its closest alarm to determine the safe region or safe period to move without checking alarms. Surprisingly, no existing research has tried to capitalize on spatial and mobility constraints of mobile devices traveling on road networks for optimizing and scaling spatial alarm services.…”

Distributed Processing of Spatial Alarms: A Safe Region-Based Approach

“…The second experiment evaluates the bitmap encoded safe region (BSR) approaches, namely grid bitmap encoded safe region (GBSR) and pyramid bitmap encoded safe region (PBSR). The final experiment provides an evaluation of the safe region techniques compared to periodic processing (PRD), safe period-based (SP) computation [3] and the optimal (OPT) approach as described in beginning of Section 4. The optimal approach does not consider any restrictions on resource availability and assumes all relevant alarms within the grid cell are pushed to the client, which implies the client is fully aware of all relevant alarms in its vicinity.…”

“…Spatial alarm processing requires meeting two demanding objectives: high accuracy, which ensures zero or very low alarm misses, and high scalability, which requires highly efficient processing of spatial alarms. Existing research on spatial alarm processing has been focused on either client-centric [2] or servercentric [3] architectures. A client-centric architecture is limited to supporting only private alarms on static target with moving subscriber or on moving target with static subscriber.…”

“…Though periodic evaluation is simple, it can be extremely inefficient due to frequent alarm evaluation and the high rate of irrelevant evaluations. The safe period-based approach [3] allows the system to overcome the deficiencies of periodic evaluation by adaptively computing a safe period for each mobile subscriber; no alarm processing needs to be performed for the mobile user before its safe period expires. However, safe period computation heavily relies on future motion estimation of the mobile user.…”