CPU Scheduling In Xen Based Virtualized Environment

Chitra N (07C21), Karthik G R (07C50), Sujitha P (07C113)

The credit scheduler is a proportional fair share scheduler that automatically load balances guest VCPUs across all vailable physical CPUs on Symmetric Multi-Processing (SMP) host. The administrator does not need to manually pin VCPUs to load balance the system. However, one can restrict which CPUs a particular VCPU may run on using the generic vcpu-pin interface. Credit based scheduler provide way to tune the per Virtual Machine guest scheduler parameters. Though network intensity is fluctuating parameter, it is vital for system response.The network awareness is integrated in credit based scheduler, which would enable better CPU utilization. Global load balancing feature of credit scheduler makes this approach an efficient one. Server load is balanced effectively with the help of scheduling in virtualized environment by taking into consideration the network communication. Our experimental evaluation with realistic web server applications and benchmarks demonstrates the performance benefits and scalability of the proposed scheme.

Meeting the challenges of HDFS Scalability

Manoj P (07C60), Yuhendar B(07C130), Gautham S (07C32)

Hadoop Distributed File System(HDFS) is a distributed file system designed to run on commodity hardware.It is highly fault tolerant.But, in terms of scalability, a few issues are to be handled.HDFS does not provide high availability because the size of the metadata increases and it consumes more RAM space if large number of small files are stored.Name Node is a single point of failure in HDFS.Also, HDFS does not provide archiving feature.The objective of the project is to make HDFS more scalable by efficiently storing large number of small files in HDFS and managing metadata efficiently.A Secondary Name Node is used to avoid single point failure.

Secure Framework for Online Voting

Hari Prasad M(07C37), Ilayaraja M(07C39), Krishnan M(07C55)

With the rapid development of internet, every day-to-day activities such as shopping, banking, trading are made online on the web. There is no exception to the migration of traditional paper ballot voting and electronic machine voting to online voting systems. On the other hand, since these applications are available to any one on the internet, security has become an important concern. The proposed project focuses on improving the security aspects of online voting system. The proposed framework includes security aspects such as privacy, anonymity, and prevents application layer attacks and network layer attacks such as intrusion and Distributed Denial Of Service (DDOS) attacks.

Internals Module in TCENet

Ilayaraja M(07C39), Krishnan M(07C55), Chitra N (07C21), Karthik G R (07C50), Sujitha P (07C113)

The Internals Module in TCENet has Student,Staff,Admin,office these users.Staff will enter the Student marks(for 50 or 100) and these marks will be converted according to student sem and their specified patterns.If the freeze date for entering marks for a subject is expired means , admin will enter those subjects.The above steps carried out for Assignment,Project Review marks.At the end of each semester through Office login student ISO Report,Parents's copy , Cumulative Internal Report , Histogram and Pass percentage for various subjects are taken

Detection of attacks using Decision tree and Naive Bayes classifier

Samiya bindi C G(06C75), Rajeswari S(07C89), Sudha V(07112)

Network intrusion detection is used to distinguishes the behaviour of the network from normal.As the network attacks have increased,the intrusion detection system in necessary to secure the network.Due to large volume of secure data as well as complex and dynamic properties of intrusion behavior, optimizing the performance of intrusion detection system becomes a crucial problem.The objective of this project is to detect attacks using Decision trees with pruning modules.And Naive Bayes classifiers with Multi Variate Bernoulli module.A subset of kddcup99 data set is used for evaluating the detection performance of the algorithm.Correlation based feature selection is used to select the most important features of the data set.The classification accuracy of Decision tree with pruning module is found better than Naive Bayes classifier in the overall classification accuracy of 99.49% is obtained from the decision trees.

An Integrated Approach for Detection and Prevention of DDoS Attacks using Enhanced Support Vector Machines and Filtering Mechanisms

Parameswaran P(07C72), Parthiban C(07C74).

Distributed Denial of Service (DDoS) attacks currently bring a tremendous threat to the information infrastructure. In DDoS attack, multiple malicious hosts that are recruited by the attackers launch a coordinate attack against one host or network victim, which cause denial of service to legitimate users. The existing technique suffers from more number of false alarms and more human intervention for attack detection. The on-line monitoring system makes the user convenient to distinguish suspicious traffic from legitimate traffic, dealing with botnets, managing and defending against DDoS attacks. The objective of this paper is to monitor the network on-line which initiates detection mechanism if there is any suspicious activity and also defense the hosts from being arrived at the network. DDoS attack detection is carried out by Enhanced Support Vector Machine (ESVM). Lanchester Linear Law is used to calculate the attack strength and defense strength. Spoofed IP's are detected using Hop Count Filtering (HCF) and the attack strength is calculated. Based on the calculated attack strength any of the defense schemes such as Rate based limiting or History based IP filtering is automatically initiated to drop the packets from the suspected IP. The integrated on-line monitoring approach for detection and defense of DDoS attacks is deployed in an experimental testbed. The on-line approach is found to be obvious in the field of integrated DDoS Detection and Defense.

Enhancing Search Engine Reliability By Crowd Sourcing User's Feedback

Azhagu Selvan SP(07C18), Dinesh Kumar M(07C26), Arun Joshi V(07C12).

The Majority of current web search engines use keyword-based method for similarity computing,they cannot discriminate important web pages among huge amount of search results. Here we propose user Feedback feature for web search engine. The proposed system will accumulate the user Feedback of the search results with the help of a browser addon and stores it in a custom, central server. Those feedback are displayed to the other users who use the same query for searching, thus increasing the search engine reliability. For this a custom server is developed with django and django-piston. A greasemonkey user script is developed for client side. The system is named a RankGoogle.

Distributed Indexing and Searching Of Large Dataset Using Map-Reduce Framework

Balagi K(09LC01), Sriram R (08C112), Vijayraajaa G S (08C127)

The project mainly aims in indexing large dataset and effectively retrieving them in a distributed fashion. Initially a large volume of data is stored in Hadoop Distributed File System and the data is preprocessed to reduce the size of the data by removing stop words. Grammatically similar words are removed by stemming process. The preprocessing step thus reduces the size by at most 1/8th of original dataset size. Then the data is indexed by the inverted index process in a distributed fashion over number of available nodes in the cluster environment. After the indexing turn, the indexed data is successfully grouped and stored collectively in HBase, which is capable of handling multiple requests at a time. With data inserted in HBase searching a term and its corresponding occurance can be done effectively.

Parallel Implementation Of Decision Tree In Hadoop Environment

Karthikeyan T(09LC04), Sabareesh R(08C90)

Building a decision tree classifier generally includes two stages, a growing stage and pruning stage. Assume that , there are three datasets on instances set D={t1,t2,...tn} an attributes set X={x1,x2,...xm} and a class label set or goal set C={c1,c2,...ci}. Constructing a decision tree from the attributes set to classify the instances set based on the class label set.

Map-Reduce Approach For A Scalable Topic-Based Search Engine

Varadharajan M

Trivial information retrieval tasks like indexing and searching on web data are becoming more complex as the amount of data on web grows exponentially. Search engines need to crawl and index billions of web pages and thereby enabling them to be retrieved for intended queries. The main objective of this project is to design and implement a scalable topic based search engine , capable of indexing and querying billions of web pages. This project makes extensive use of 'MapReduce' and Latent Dirichlet Allocation to cluster documents together on the basis of latent topics.

Video Surveillance Using Cloud Computing
Someshwar B C , Sasikumar T.

Video surveillance is finding its application in various areas like crowd management, traffic automation, theft detection. The automation of this surveillance process is becoming a major issue. Video surveillance involves complex image processing techniques that are highly computation intensive. Being a real time system computational delay is not accepted. A lot of image data has to be stored which requires a huge storage. All these factors require huge infrastructure and dedicated server which is costly. The main aim of this project is to provide and automated Video surveillance process as a service in the cloud thereby eradicating need for investing money on dedicated servers. The algorithm is a parallelized using message passing queues. The video images are saved in a compressed form thereby saving space. The proposed methodology gives efficient results for complex image processing applications.

Detecting Intrusions In Virtualized Environment

Ramya M V, Sivananthinikumari C, Sumathi M

Intrusion detection system has been introduced and broadly applied to the prevent unauthorized access to the system resource and data for several years. However, many problems are still not well resolved in most of IDS, such as detection evasion, intrusion containment. In order to resolve these problems, we propose a novel flexible architecture using Virtual environment which is based on virtual machine monitor and has no-intrusive behavior to target system
after studying popular architecture. In this architecture a separate IDD is added to provide intrusion detection services for all virtual machines. IDD, as the core component of IDS, is separately isolated from the target systems, so strong reliability is also achieved in this architecture. And we have compared the various tools to select one tool to be used in IDD. The tools are compared by their execution time, types of malware detected the platforms they support and their effectiveness.

Performing Live Migration In Virtualized Environment

Thanumoorhty N, Ramlin R, Shyam Sundar S

Cloud computing is the hottest topic in IT industry that enables outsourcing computations on demand. The efficiency of cloud computing is improved by creating Virtualized environment which gives the benefit of flexible infrastructure, high availability etc. But security is the major concern in virtualized environment that affects benefits like high availability of cloud computing. In such case the cloud can provide service to the requestors because of attacks. To overcome this, the concept of live migration is used in the proposed method where any attacks like
DDOS happen in physical hosts where virtual machines are providing its service, then the virtual machine is automatically live migrated to another physical host where it continues to provide its service uninterruptedly. Thereby the cloud is made available despite of attacks.

Handling Multiple Jobs Using Triple Queue Scheduler In Cluster Environment

Raghavendhra B, Saravanan P S

Map Reduce: a programming model and an associated implementation for processing and generating large sets. In a practical data center of that scale, it is a common case that I/O- bound jobs and CPU-bound jobs, which demands different resource run simultaneously in the same cluster. In the Map Reduce framework, parallelization of these two kinds of jobs has not been concerned. In this project a new view of the Map Reduce model is proposed through classification of the Map-Reduce workloads into two categories based on their CPU and I/O utilization and scheduling them. With workload classification, a Triple Queue schedule is designed, which detects the workload type on the fly and schedule multiple job at a time. The Triple-Queue schedule would improve the usage of both CPU and disk I/O resource under heterogeneous workloads.