The feasibility of practical in-band full-duplex radios has recently been demonstrated experimentally. One way to leverage full-duplex in a network setting is to enable three-node full-duplex, where a full-duplex access point (AP) transmits data to one node yet simultaneously receives data from another node. Such three-node full-duplex communication, however, introduces inter-client interference, directly impacting the full-duplex gain. It hence may not always be beneficial to enable three-node full-duplex transmissions. In this paper, we present a distributed full-duplex medium access control (MAC) protocol that allows an AP to adaptively switch between full-duplex and half-duplex modes. We formulate a model that determines the probabilities of full-duplex and half-duplex access so as to maximize the expected network throughput. A MAC protocol is further proposed to enable the AP and clients to contend for either full-duplex or half-duplex transmissions based on their assigned probabilities in a distributed way. Our evaluation shows that, by combining the advantages of centralized probabilistic scheduling and distributed random access, our design improves the overall throughput by 1.53 times, on average, as compared with the greedy downlink-uplink client pairing.

The feasibility of practical in-band full-duplex radios has recently been demonstrated experimentally. One way to leverage full-duplex in a network setting is to enable three-node full-duplex, where a full-duplex access point (AP) transmits data to one node yet simultaneously receives data from another node. Such three-node full-duplex communication however introduces inter-client interference, directly impacting the full-duplex gain. It hence may not always be beneficial to enable three-node full-duplex transmissions. In this paper, we present a distributed full-duplex medium access control (MAC) protocol that allows an AP to adaptively switch between full-duplex and half-duplex modes. We formulate a model that determines the probabilities of full-duplex and half-duplex access so as to maximize the expected network throughput. A MAC protocol is further proposed to enable the AP and clients to contend for either full-duplex or half-duplex transmissions based on their assigned probabilities in a distributed way. Our evaluation shows that, by combining the advantages of centralized probabilistic scheduling and distributed random access, our design improves the overall throughput by 3.16× and 1.44×, on average, as compared to half-duplex 802.11 and greedy downlink-uplink client pairing.

Large data analysis is an important topic in cloud computing. Large-scale data analysis requires complex data analysis, such as Theta-Join, which includes equi-join and nonequi-join. On the other hand, MapReduce is a programming framework in cloud computing to compute data analysis in parallel. In order to improve MapRduce performance in complex data analysis, researchers propose the Map-Join-Reduce API to support the equi-join operation. The proposed method not only extends the Map-Join-Reduce framework but also supports nonequi-join. We propose three concepts. First data are filtered first according to the query statements. Second, the filtered data are sent to its corresponding worker according to the join expression for higher level parallelism. Each worker then performs the corresponding join operation after receiving the filtered data. Finally, we aggregate the result by using aggregate functions specified in the select clause.

In hard-real-time environment, scheduling periodic tasks upon multiprocessors is one of the most popular problems where uniform multiprocessor scheduling is a well-known one. In this platform, execution time of each task in one processor is proportional to the computing capacity of this processor. From previous works, we know there are only approximate feasible solutions for on-line scheduling on uniform multiprocessors. In this paper, with task migration, we first present a novel model called T-Ler plane to describe the behavior of tasks and processors, and two optimal on-line algorithms based on T-Ler plane to schedule real-time tasks with dynamic-priority assignment on uniform multiprocessors. To make it practical and to reduce context switches, we also present a polynomial-time algorithm to bound the times of rescheduling in a T-Ler plane. Since task migration is easier in SOC multicore processors, our result might be applicable and adapted to many asymmetric multicore platforms.

XML has become the most important technique to exchange data in WWW. Providing efficient access to XML document databases is thus crucial. Concurrency control protocols allow transactions to be executed concurrently to improve performance. However no proposed research aims at the concurrency control protocol for XML databases. In this paper we propose a lock-based concurrency control protocol, called XLP for XML documents. XLP takes into consideration the access behavior of the XPath model, which is an important language for addressing data in XML documents suggested by W3C XLP has the features of richer lock modes, refined lock granularity, lower lock conflict and lock conversion. We prove that XLP always generates serializable schedules. Our simulation results show that, for XML documents of different sizes, XLP outperforms 2PL by 84.4% on average and outperforms the tree locking protocol by 95.6% on average. The performance comparison of XLP with both 2PL and the tree locking protocol is also made in this study for various transactions with different XPath lengths, read/write ratios and the existence of predicates.