We report on design, implementation and evaluation of a resource management system that builds upon OpenStack, an open-source cloud platform for private and public clouds. Our implementation supports an Infrastructure-as-a-Service (IaaS) cloud and currently provides allocation for computational resources in support of both interactive and computationally intensive applications. The design supports an extensible set of management objectives between which the system can switch at runtime. We demonstrate through examples how management objectives related to load-balancing and energy efficiency can be mapped onto the controllers of the resource allocation subsystem, which attempts to achieve an activated management objective at all times. The design is extensible in the sense that additional objectives can be introduced by providing instantiations for generic functions in the controllers. Our implementation monitors the fulfillment of the relevant management metrics in real time. Testbed evaluation demonstrates the effectiveness of our approach in a dynamic environment. It further illustrates the trade-off between closely meeting a specific management objective and the associated cost of VM live-migration.

/pdf/dynamic-resource-allocation-with-management-objectives-2bz4zu98or.pdf

Dynamic resource allocation with management objectives: implementation for an OpenStack cloud

Sentiment Analysis (SA) is an application of Natural Language Processing (NLP) to extract the sentiments expressed in the text. In this paper, we experimented five approaches to perform SA, namely, Lexicon based approach, Supervised Machine learning based approach, Hybrid approach, K-means with Bag of Word (BoW) approach and K-modes with BoW approach. We have experimented these approaches using five corpora with different feature representation techniques to predict the best approach to perform SA in Tamil texts. In this research we used Basic features such as word count and punctuation count in addition to traditional features such as Bag of Words (BoW) and Term Frequency-Inverse Document Frequency (TF-IDF) included to check their influence in the prediction. We have compared these approaches, features and the corpora. From the evaluation the highest accuracy of 79% is obtained for UJ_Corpus_Opinions_Nouns corpus with fastText for supervised Machine learning based approach.

Sentiment Analysis in Tamil Texts: A Study on Machine Learning Techniques and Feature Representation

A case study and critical assessment in calculating power usage effectiveness for a data centre

Unlike a traditional virtual machine (VM), a container is an emerging lightweight virtualization technology that operates at the operating system level to encapsulate a task and its library dependencies for execution. The Container as a Service (CaaS) strategy is gaining in popularity and is likely to become a prominent type of cloud service model. Placing container instances on virtual machine instances is a classical scheduling problem. Previous research has focused separately on either virtual machine placement on physical machines (PMs) or container, or only tasks without containerization, placement on virtual machines. However, this approach leads to underutilized or overutilized PMs as well as underutilized or overutilized VMs. Thus, there is a growing research interest in developing a container placement algorithm that considers the utilization of both instantiated VMs and used PMs simultaneously. The goal of this study is to improve resource utilization, in terms of number of CPU cores and memory size for both VMs and PMs, and to minimize the number of instantiated VMs and active PMs in a cloud environment. The proposed placement architecture employs scheduling heuristics, namely, Best Fit (BF) and Max Fit (MF), based on a fitness function that simultaneously evaluates the remaining resource waste of both PMs and VMs. In addition, another meta-heuristic placement algorithm is proposed that uses Ant Colony Optimization based on Best Fit (ACO-BF) with the proposed fitness function. Experimental results show that the proposed ACO-BF placement algorithm outperforms the BF and MF heuristics and maintains significant improvement of the resource utilization of both VMs and PMs.

/pdf/a-placement-architecture-for-a-container-as-a-service-caas-4e83074n18.pdf

A placement architecture for a container as a service (CaaS) in a cloud environment

The significant role of long non-coding RNAs (lncRNAs) in various cellular functions, such as gene imprinting, immune response, embryonic pluripotency, tumorogenesis, and genetic regulations, has been widely studied and reported in recent years. Several experimental and computational methods involving genome-wide search and screenings of ncRNAs are being proposed utilizing sequence features-length, occurrence, and composition of bases with various limitations. The proposed classifier, Deep Neural Network (DNN) is fast and an accurate alternative for the identification of lncRNAs as compared to other existing classifiers. The information content stored in k-mer pattern has been used as a sole feature for the DNN classifier using manually annotated training datasets from LNCipedia and RefSeq database, obtaining accuracy of 98.07 %, sensitivity of 98.98 %, and specificity of 97.19 %, respectively, on test dataset. The k-mer information content generated on the basis of Shannon entropy function has resulted in improved classifier accuracy. This classification framework was also tested on known human genome dataset, and the framework has successfully identified known lncRNAs with 99 % accuracy rate. The said algorithm has been implemented as a web prediction tool, which is available on server interface http://bioserver.iiita.ac.in/deeplnc.

DeepLNC, a long non-coding RNA prediction tool using deep neural network

In today's IT computing era, there is shift to aggregate computing resources into data centers (DC) shared by pool of users. With the increasing popularity of cloud computing, this pool of users are now more global than any-time in the past The resource demand of these users are inadvertently governed by strict “service level agreements”. In this paradigm, data centers' operational energy costs are a rising concern as they continue an upward trend that is poised to surpass the capital cost of equipment in a typical lifetime usage model. A data center is a complex distributed system comprised of a hierarchy of numerous components; and operating at multitude levels of usage abstractions. Virtualization is an approach to consolidating multiple services onto a smaller number of computing resources. A virtualized server environment allows computing resources to be shared among multiple performance-isolated platforms called virtual machines. Virtual Machine(VM) technology has been widely applied in data center environments due to its seminal features, including reliability, flexibility, and the ease of management. We define set of service-provider controllable policies comprising of number of physical servers, possible number of VMs, possible VM request queue length and maximum number of requests per VM threshold. For each of these defined policy, we implement a dynamic resource provisioning framework for virtualized server environment which achieves power efficiency by using optimum power efficient allocation and workload forecasting scheme to ensure that workload request meet the defined SLAs and the data center operates with least power consumption. Finally, based on the results from our implementation, we analyze and infer on the best policy which meets SLA commitments and has the least energy consumption.

Power aware provisioning in cloud computing environment

Power management in virtualized datacenter - A survey

Growth of Cloud computing has fueled the demand for large infrastructures called data centers. While the objective of the cloud computing ecosystem is to move computation and processing away towards a centralized environment, the fallout lies in the power and energy consumption of these infrastructures. Reducing power consumption is an essential requirement for Cloud resource providers to decrease operating costs. One of the options to reduce power consumption is to reduce the number of servers in IDLE (unused) state— as these IDLE servers consume as much as 60% of peak power. Number of servers in IDLE state can be reduced by turning off IDLE servers or transitioning these IDLE servers to low power SLEEP state. With virtualization being the backbone to provision cloud computing services, we use simulation to study and report the impact of using SLEEP state on the server with its virtual machines (VM) servicing application workload requests. We look at two parameters: a) power consumption of the cloud computing environment, b) average response time per request. Our simulation results show that using SLEEP state at server level, and the server with its VMs' servicing application workload requests, we can achieve a 2% savings in average power usage and around 27% savings in average response time per request.

A study on server Sleep state transition to reduce power consumption in a virtualized server cluster environment

Data mining is the process of extracting interesting patterns or knowledge from large amount of data. With the development of data mining technology, an increasing number of data can be mined out to reveal some potential information about the user, because of which privacy of the user may be violated easily. Privacy Preserving Data Mining (PPDM) is used to mine the potential valuable knowledge without revealing the personal information of the individuals. K-Anonymity is one of the Privacy Preserving model that aims at making the individual record be indistinguishable among a group records by using techniques of generalization and suppression. The existing approaches are based on homogeneous anonymization that anonymizes quasi attributes by choosing a single sensitive attribute. This approach causes high information loss and reduces the data utility. To overcome these issues in the existing system, sensitive attribute based non-homogeneous anonymization system is proposed. Based on the sensitive attribute, non-homogeneous anonymization technique (generalization and suppression) is applied to the identified quasi attributes and the non sensitive attributes are directly published. Thus the proposed system achieves high degree of data utility, reduces information loss and also achieves high degree of Data Integrity.

Sensitive attribute based non-homogeneous anonymization for privacy preserving data mining

The emergence of small, resource-constrained, and highly mobile computing platforms presents numerous challenges in developing and providing new services for users on the move. These challenges demand architectures and infrastructures that are scalable, reliable, secure and personalised. Recently, middleware has emerged as one of the key solutions for supporting next generation ubiquitous communication, and providing additional personalised services over and above those provided by existing networks. To this end, this paper presents the design of a community network-based middleware system for providing personalised information access to mobile users. The paper also discusses a prototype that has been implemented to validate and demonstrate the features of the middleware. Initial results regarding usability, flexibility, transcoding, memory efficiency and display rendering efficiency are discussed. Future work for the mobile middleware and research directions are outlined.

R. Shriram

Papers

Power aware provisioning in cloud computing environment

Power management in virtualized datacenter - A survey

A study on server Sleep state transition to reduce power consumption in a virtualized server cluster environment

Sensitive attribute based non-homogeneous anonymization for privacy preserving data mining

A middleware for information processing in mobile computing platforms