An efficient multipath transport is an emerging issue because it provides greater aggregated throughput than the usage of one active path. An extension of the stream control transmission protocol that serves the multipath transport is the concurrent multipath transfer (CMT-SCTP). Using the CMT-SCTP over dissimilar paths is a challenge. Two paths are dissimilar if they possess different characteristics such as bandwidth or delay. Moreover, CMT-SCTP maintains a common buffer at the sender and receiver. No matter how the paths differ in quality, the fast paths are unable in improving aggregated throughput because of the out-of-sequence data at the receiver. This out-of-sequence data may also cause buffer blocking. The current state-of-the-art technique suggests that, the buffer should be split through various paths, in accordance with the number of paths. The technique assigns an equal amount of buffer space to the slow and fast paths. This also limits the performance of fast paths because of the small parts of the buffer. We propose a buffer splitting technique (CMT-RTTA) that maintains the outstanding data over the paths based on RTT (round trip time). A destination with a shorter RTT will be allowed to occupy more buffer space than those with a longer RTT. Additionally, we adopted a priority where the sender transmits data to the destinations with shorter RTTs first, followed by the destinations with longer RTTs. To evaluate the performance of CMT-RTTA we have established a real Internet lab setup. The experiments show that CMT-RTTA improved the throughput of CMT-SCTP on average by 14% when the maximum bandwidth dissimilarity is applied.

Concurrent Multipath Transfer (CMT-SCTP) using the Stream Control Transmission Protocol (SCTP) over multiple paths can achieve an aggregated transmission throughput much greater than a single path. A problem arises when the fair round robin scheduling (RRS) in CMT-SCTP allows the slow paths to affect the overall throughput due to out-of-sequence data at the receiver. Data along the slow and fast paths occupy the shared receiver buffer while waiting for the out-of-sequence data, which can easily lead to performance degradation. Therefore, intelligent scheduling of data transmissions is necessary for efficient data transfer, particularly in realistic situations where the paths are dissimilar and the buffer size is limited. We propose an algorithm for scheduling data transmissions based mainly on the outstanding bytes (CMT-OUT). The algorithm updates the path quality after a successful transmission, which is a measure of how preferable the current destination would be when selection happens in the next round. The experiments on a realistic testbed show that CMT-OUT improved the throughput of CMT-SCTP on average by 21% when the maximum bandwidth dissimilarity is applied over a two-path network. The mechanism in CMT-SCTP to handle the delay dissimilarity in a simple two-path scenario is very efficient due to the rare timer based retransmissions. In the experiments for the more complex four-path scenario, CMT-OUT improved the throughput by 54%.

Concurrent multipath transfer using the stream control transmission protocol (CMT-SCTP) enables multi-interface devices to send data simultaneously over multiple paths. The multipath transfer requires special handling in real life practice due to the different path characteristics, such as a path using 4G and a path using Wi-Fi. Moreover, for smartphones and other portable devices, the resources, such as the sender and receiver buffer are limited due to the short battery life. The buffer resource is common for all paths when working with the CMT-SCTP. In this paper, we have evaluated the performance of CMT-SCTP under the dissimilarity between the paths caused by the propagation delay with a common and a split buffer. We have applied the buffer splitting (CMT-RTTA) based on the round trip time (RTT), so that the paths with longer RTT (slow paths) occupy less buffer resource both at sender and receiver. In order to enhance the transmission throughput, we have also scheduled the transmissions based on the RTT value to the corresponding destination address. For the evaluation, the experiments are performed on a two-path and a four-path network. CMT-RTTA performs as good as the basic CMT-SCTP over a simple two-path setup, whereas, on a complex four-path setup, CMT-RTTA outperforms the CMT-SCTP.

In this study we implement CMT-SCTP (Concurrent Multipath Transfer extension of the Stream Control Transmission Protocol) in a real lab Internet setup. SCTP is an innovative transport layer protocol, which is reliable and provides services such as multihoming and multistreaming. CMT extension allows simultaneous utilization of multipaths for reaching the higher data transmission rate between two hosts, which is beneficial for the emerging multi-interface devices. The literature on CMT-SCTP optimization shows that, there are several simulation tools used by the researchers for performance evaluation, to validate propositions and to experiment the further extensions of SCTP/CMT-SCTP. Indeed, such tools are convenient and can quickly verify the proposal results. We establish a real lab Internet setup first and then design a number of test scenarios for CMT-SCTP. To verify our implementation the simulations needs to be performed on the test scenarios to set a reference output prior to the lab setup experiments. The experiment results from the successful implementation of CMT-SCTP on the real lab setup agrees to the reference output reasonably.

Automatic identification and classification of human actions is one the important and challenging tasks in the field of computer vision that has appealed many researchers since last two decays. It has wide range of applications such as security and surveillance, sports analysis, video analysis, human computer interaction, health care, autonomous vehicles and robotic. In this paper we developed and trained a VGG19 based CNN-RNN deep learning model using transfer learning for classification or prediction of actions and its performance is evaluated on two public actions datasets; KTH and UCF11. The models achieved significant accuracies on these datasets that are equal to 90% and 95% respectively on KTH and UCF11 which beats some of the accuracies achieved by handcraftedfeature based and deep learning based methods on these datasets.

With easy access and many services such as social networks, online shopping, video streaming the data traffic over the Internet is increasing. On the other side, the traditional congestion control strategies of TCP due to the huge data are not sufficient. The TCP protocol uses such traditional techniques to minimize the network congestion. Moreover, handling applications with smartphones is challenging in terms of congestion due to the long delay networks such as 4G. Many TCP variants have been proposed for the network congestion particularly for long delay networks such as TCP (Binary increase congestion control) BIC and CUBIC. These proposed techniques for TCP are also not perfect and they require extensive experimentation over long delay networks. In this research, the performance evaluation is accomplished between three variants of TCP for network congestion i.e., TCP, BIC and CUBIC. Particularly the simulations are proposed in order to trace the throughput and fairness of these congestion control techniques. Over a number of simulations it is concluded that under the proposed topology TCP CUBIC improves the throughput and increases the bandwidth fairness approximately by 38% when compared with the basic TCP.