4.3 Testing the Sensor–Mobile Communication
4.3.3 Error and Request Re–Assignment
In the last experimental setup, we evaluate the reliability of each flow pattern in terms of accessing data in a group sharing environment when there is failure. Since errors are bound in wireless communication mediums (e.g., Bluetooth), sensor data (e.g., readings from the T1 Sensor Tag) will experience delayed responses or failures. Also, there is no guarantee that communication between the adjacent edges will be seamless since users are mobile within a shared-space.
In this experiment, we employed with the fault–injection technique where we evaluate specifically, the accuracy of re–assignment of a transaction to the next desired node. We define transaction within the concept of retrieving sensor data from a mobile host. With fault injection, the idea is to request a data and ”kill” the intended host so that another host within the discoverable list of devices can be contacted. We carefully keep terminating the transactions unexpectedly until at least 5 iterations can be made before the desired sensor information is retrieved. The result is presented in Fig. 4.14 and Fig 4.15.
Figure 4.14: Correctness of Task Re–Assignment
In Fig. 4.14, we studied the correctness of re–assigning a task to the next desired node following the underlying principle of each flow pattern. We found that the parallelism flow has better accuracy with the task assignment than the other two flow patterns. A reason for this observation is that, the parallelism approach focuses only on the responses since the requests are concurrent. The flow pattern only considers the responses and their order of arrival. Practically, the first response with the desired data is selected so
there is only little room for error. The Choice flow pattern however outperforms the sequential flow in terms of correctly assigning tasks. This happened because the choice flow ignores the processing time of the mobile host (TP) and only focuses on the request-response travel time (RTT). This means, the algorithm focuses on only one factor and the chances of identifying/determining a smaller RTT over another is higher than doing extra calculations.
Figure 4.15: Tasks Assignment to Wrongly Identified Mobile Host
The sequential flow sees its correctness value dropping as the fault increases because we are comparing two factors (the RTT and PT) and the number of comparisons are doubled. Another reason is because the PT factor of the host device is not stable and changes can occur between the time a request is issued and a response is received. Future works have to focus deeper on addressing the dynamism of PTs in the entire architecture.
In both Fig. 4.14 and Fig. 4.15, it is observed that the flow patterns that are least in terms of correctly re–assigning a task hive higher error rate in terms of wrong task assignment.
4.3.4
Summary
The concept of mobile services hosting is gaining more attention most recently due to the expansion and diversity of macro fields such as Internet of Things (IoT) and Cyber–physical systems. With mobile hosting, services (including data and application states) can be provisioned from mobile devices such as smartphones and tablets; emulating actual servers. This can greatly improve on mobile–to–mobile communication. More- over, other forms of data (e.g., sensor data) can be collected from sensor devices and sent to the mobile host to be made available to other devices.
cyber–physical systems require group sharing since sensor information can be gathered from heterogeneous sources. The problem in such situations however is the ability to reduce latency through the minimization of communication overhead. Latency reduction is critical because delays in the accessibility of the services in the mobile hosting and sensor system can lead to undesired situations such as information misreading, failed request delivery, wrong ordering of responses, and so on.
This work explores the area of mobile services hosting and the ability to share sensor information espe- cially in a group–sharing scenario. The paper adapted the concept of edg–based services accessibility from distributed cloud computing to create the mobile hosting environment. However, the work advanced on the area by proposing the idea of accessing the services from an adjacent node with the minimal time that considers the total travel time of the request and response (RTT) plus the processing time (PT) of the host. This is the typical scenario that the CSB–UCC is advocating for. Existing works only consider the most optimal travel time of the request between the adjacent nodes on the edge. Furthermore, we studied varied transactional flow patterns that can facilitate better performance regarding the minimization of latency.
The work is evaluated to determine the best approach for achieving low–latency communication and efficient job re–assignment. The preliminary evaluations show that the proposed consideration for the optimal RTT + PT is better than the existing approaches that evaluate latency solely on the optimal RTT. Also, the results show that the parallelism flow pattern is better than the other two which are the sequential and choice flow patterns. It is important to state that this work in conjunction with the Environmental Instruments Canada Inc. adopted the best–proximity technique for the mobile–sensor–data sharing.
In summary, the contributions in regards to the support of sensor communication of the proposed CSB– UCC includes:
• Proposed mobile hosting architecture for group data sharing. Existing works only focus on the feasibility of the idea while enterprise grade applications on mobile provisioning are few. Existing techniques including the SOPHRA [201] framework do not support sensors.
• Proposed and evaluate different communication flow patterns based on sequential, parallelism, loop, and choice methodologies.
• It is observed that optimal time for a response is not dependent on optimal distance between the adjacent mobile and sensor nodes but factors such as the processing load on the host, and request travel time should be collectively considered.
• Failed communications can be re–routed to the adjacent node that has the next better optimal request– response time.
In most of the scenarios, the parallelism flow pattern is better at latency minimization. There is however more room for improvement on the current state of the work. First, there is the need to study more rigorous error management techniques when communications between the mobile host and its consumer
fails. In a mobile hosting environment, errors can be introduced at two levels, communication failure and system failure on the host. Therefore, the two can be studied in detail in another work. Another issue for future consideration is energy conservation on the mobile devices since most applications in mobile hosting environment run constantly either in the foreground or background.
In the next section, the attention of the dissertation will switch to scalability testing of the CSB–UCC. This is another major consideration of the dissertation.