Sheng Wei Huang

Improving Speculative Execution Performance with Coworker for Cloud Computing

MapReduce is an important programming model for large-scale parallel applications. It divides a job into several parallel tasks and completes the job by sequential phases, i.e. map phase and reduce phase. The job completion time will be delayed when a task, called straggler, consumes more time than others. The main reason that a straggler occurs is the imbalance resource distribution among computing nodes in the cloud. Speculative execution is a solution for dealing with stragglers. Duplicate tasks are launched on other nodes to process the same data as the straggler does. Any completion of these tasks implies that this task is finished and other duplicate tasks can be aborted. However, aborting tasks misspends resources. In this paper, we propose an idea of using coworkers to help a straggler. According to the processing rate of the straggler and the coworker, the amount of data parceled out from the straggler to the coworker should be determined. Different from speculative execution, coworkers finish tasks with stragglers and do not misspend computing resources. Experimental results show that coworkers can reduce the task completion time by 37% and the network traffic by 64% when comparing with speculative execution.

A topology-based scaling mechanism for Apache Storm

Summary As more and more well-known companies, such as Twitter, Yahoo, and Alibaba, start to focus on real-time big data applications, how to build a platform for processing real-time data becomes an important issue. Among all the real-time processing systems, Apache Storm is the most well-known and representative open-source, distributed, real-time computation system. In Storm, the computation is implemented by a topology such as a graph where nodes are operators and edges represent the data flows between operators. In big data processing and analysis systems, scalability is an important issue. Storm provides rebalance mechanism for its scalability property, which can adjust the parallelism of a running topology. However, there are some drawbacks in rebalance command, such as resource usage restriction and topology execution suspension. In this paper, we propose a topology-based scaling mechanism for Apache Storm. When a topology is overloaded, it scales by adjusting the number of the cloned topologies or replaced by another new topology with more tasks. When scaling by topology-based mechanism, it eliminates resource usage restriction and execution suspension in the topology, and the procedure is automatically launched. The experimental results show that our topology-based scaling mechanism can improve the scaling performance of Storm. Copyright

IPv6 operations and deployment scenarios over SDN

IPv6 is a technology that provides enormous address space and end-to-end communication, features that are required in the context of the device automation integration for future network. The transition to IPv6 holds the future of the internet infrastructure. Software-defined networking (SDN) defines a new concept for computer networks that can separate and provide abstract elements of network devices. IPv6 SDN has the potential to revolutionize the network design, construct and operate networks to achieve more efficient business network agility. In this paper, we will discuss the main architectures of SDN and illustrate how IPv6 can be deployed and integrated in SDN technologies using OpenFlow mechanisms. We will also discuss the IPv6 impact on link performance and deployment scenarios.

Implementation of virtual network management system with SLA on NetFPGA

With the increase in the bandwidth needs of network service, the customer requirements for Quality of Service (QoS) are also higher. Although the QoS guarantees can be offered by signing the Service Level Agreement (SLA), the costs of infrastructures are increased as well. This paper proposes a shortest path algorithm based on SLA, combined with the resource allocation for virtual networks (VNs). It can help service providers to enhance the achievement of overall SLA contracts in the case of lack of resources. This architecture was also implemented on NetFPGA platforms applied with the concept of OpenFlow to meet the demands of network performance and flexibility. Finally, the experimental results show that the proposed algorithm can dynamically provide the better path to reach the SLA achievement rates.

A Region-Based Allocation Approach for Page-Based Scratch-Pad Memory in Embedded Systems

In this paper, we propose a new dynamic approachfor addressing the allocation problem of the pagebasedscratch-pad memory (SPM) in embeddedsystems. The main characteristic of the proposedapproach is to merge small basic blocks of a programinto big regions, and then perform the SPM allocationonly when the control flow of the program istransmitted from one region to another. The results ofour performance evaluation have shown that theproposed approach effectively increases the hit ratio ofthe SPM while simultaneously reduces the CPU idletime caused by the SPM reallocation. Consequently,the energy-delay-product (EDP) values of the testprograms are significantly improved by 60% inaverage.

Efficient and Scalable SPARQL Query Processing with Transformed Table

Resource Description Framework (RDF) is the core technology of Semantic Web and has been more and more popular in recent years. With the rapid growth of the RDF data, the Triple Store, which is the query engine and RDF data storage, requires more scalable and efficient technologies. To improve the scalability and the performance of triple query, which is called SPARQL query processing, Map Reduce programming model and NoSQL database system such as H Base are well-known solutions for large scale data processing. However, in general case, the subject of a triple is regarded as Row Key in the table. In some queries, finding matched triple patterns is a time-consuming job. Therefore, we design another table with different storage schema called Transformed Table to reduce the time cost for read operation. The experimental results show that using Transformed Table can improve the triple query performance significantly.

Cooperative Car Video Share and Search System

The car video can protect the driver by clarifying the liability in the event of an accident. More and more devices and functions have been on sale for drivers in recent years. They are focused on high quality images, wide angle support and night vision functions. However, those technologies cannot help to clarify the accident liability when the sight and angle are obscured by neighbouring cars or buildings. Hence, this paper proposes the Cooperative Car Video Share and Search (CCVSS) system. The car video will be uploaded frequently, including position, heading and timestamp information on each car in the CCVSS system. Hence, the driver can search for neighbouring cars’ videos when an accident happens using the above information. This paper implements a system to help the driver to upload the car video and obtain neighbouring cars’ videos via a smartphone device.Keywords—cooperative; car video; smartphone device

Automatic self-suspended task for a mapreduce system on cloud computing

A MapReduce system gradually becomes an essential technology to achieve the large scale computing on cloud computing. A MapReduce system currently is designed to distribute tasks over nodes in a cloud according to manual configurations of slot numbers in nodes. However, a MapReduce system may have the performance degradation due to the inappropriate configuration of the slot number, because the slot number can not exactly reflect the performance of the node. A MapReduce system can utilize the Automatic Self-Suspended Task (ASST) proposed in this paper to alleviate the performance degradation due to the inappropriate configuration of the slot number in a node on cloud computing. In experiments of this paper, a MapReduce system is proved to have a better performance with the help of ASST for various applications on cloud computing.

Scanning Files for Signatures with a MapReduce System on Cloud Computing

Because malicious code gradually endangers peoples life and verifying the harmlessness of a file usually costs many resources, the MapReduce System for Online File Scanning (MRSOFS) is proposed as a solution for combing the online file scanning service and a MapReduce system on cloud computing. In this paper, MRSOFS is introduced, implemented, and evaluated with experiments. In experiments, MRSOFS outperforms the standard case of using a single computer to scan a file for a specific signature in a signature database. MRSOFS is proved to get performance benefits by using a MapReduce system to scan a file for a specific signature, and can be utilized by an online file scanning service against malicious code.

A simple cluster-scaling policy for MapReduce clouds

Due to the rise of cloud computing, many cloud services have been developed. Google proposed a programming model called MapReduce for processing large amounts of data. After YAHOO! proposed Hadoop, many companies and enterprises have started using this programming model to establish their own cluster for handling large amounts of data. Computing resources within a cluster are often not all be used. Therefore, many researches about cluster-scaling are presented. These studies were proposed to reduce the size of the cluster to achieve power saving or to add more computing nodes in order to obtain better performance. However, there is always a trade-off between performance and power saving. Therefore, taking both performance and energy saving into account, we propose a simple policy which can effectively identify how many computing nodes can be inactivated from a cluster without affecting the execution time. We evaluate our policy in many cases to prove that it is well-performed in different configurations and achieves performance and power saving both.