Improved Word Prediction based on N-grams and POS-Tagging

Thirumagal.C(06C105),Sowndarya.S(06C94),Saleema Hussain Manikfan(06C74)

Word prediction is the problem of guessing which words are likely to follow in a given segment of a text to help a user with disabilities. As the user enters each letters of the required word, the system displays a list of the most probable words that could appear in that position. In our project, a word predictor with N-Grams and POS-Tagging for the English language has been designed and implemented. Three standard performance metrics were used to evaluate the system including prediction accuracy, prediction proximity and Key Stroke Saving. The proposed system achieved 42% prediction accuracy, 32% prediction proximity and 45% Key Stroke Saving which is far better than the results reported so far.

The code is available at http://wordpredictor.sourceforge.net

Distributed Change-point Detection of DDoS Attacks

Aravind.J(06C08), Venkatachalam.S(06C107), Raju.T (06C66)

Security threats for the network services has been constantly increasing day by day. Distributed Denial of Service (DDoS) attack is one such kind of security threat which involves multiple systems generating a large amount of traffic towards a target machine and thereby making any service from the target machine or server unavailable to its clients. This threat by nature needs no control over the target system. Traditional methods of detecting DDoS attacks is highly disadvantageous. To overcome the disadvantages of those schemes, we are proposing a distributed methodology which involves installing the attack detectors at various parts of the network. Each router in the network will monitor the traffic flowing through it and if any anomaly in traffic pattern is detected, it will raise an alarm to the nearby routers. The alarm propagate from the router nearer to the target to the routers nearer to the sources. By this way a tree like construct is made, which will have information about number of alarms raised and the path of attack flow. If the construct shows any converging pattern, then it is declared as DDoS attack.

Profile Driven Scheduling

Madana gopal.T(06C46), Krishna Kumar.K.B(06C45), Seethalakshmi.G.R (06C84)

Clusters of commodity servers are increasingly the platform of choice for running computationally and IO intensive jobs in a variety of industries. It is expected that using clusters will reduce the average job response time. Improper submission of jobs to clusters may lead to two severe problems. On one hand it leads to blocking of jobs(waiting for results from other jobs). On the other hand it leads to disturbing other jobs(i.e other jobs may be blocked due to submission). Effectively utilizing the resources of clusters can help to balance the load and avoid situations like slow run of systems. This project addresses the principle of effective utilization of cluster resources by profile driven scheduling.It avoids the above problems by allocating jobs to cluster based on the profiling results. Some of the jobs run effectively on machines with one set of machine configuraion parameters like CPU load etc ,while some other jobs run effectively on machines with other set of the same parameters. CPU load,free hard disk space are the parameters considered for the proposed approach. Job dependancy analysis can be used to prevent dependent jobs to keep blocking and disturbing other jobs. We have used these ideas to perform job allocation by analyzing available cluster parameters and analyzing input jobs to minimize the schedule length.

Services Based Inter-Application Authentication Framework For Tce Web Applications

Anugraha, S (06C07), Guruprasad, L(06C26), Manikandan, K(06C48)

There are functionally different web applications at Thiagarajar College of Engineering like TCENet, the intranet portal and TCE Attendance Monitoring System (TAMS), which have different code bases and different databases. The existing model introduces code redundancy especially in cases where the functionality is the same. For example, in the case of authentication, same user name, password combination is used to authenticate a user into TCENet and TAMS web applications. Therefore, same user information lying around in TAMS and TCENet databases leads to complex maintenance process. Data consistency is also affected since there are chances of changes in any one of the database. The proposed project provides a unified authentication mechanism which addresses. Different users with different access right and serves the requests from different users. Users may want access to specific services, for instance, a user may want access to TAMS but may not use TCENet at all. This framework handles such use cases robustly and provides an inter-application authentication mechanism and session sharing between the web applications.

Resource Aware Scheduling Of Map Reduce Tasks in Cluster Environment

Abhilaash, R(06C02), Arya. A.P, (06C12), Kanagu, R.M (06C39)

Task scheduling is a key element in achieving high performance from cluster computing. To be e_cient, scheduling algorithms must be based on a cost model appropriate for computing systems in use. The optimal scheduling of tasks is NP-complete and a large number of heuristic algorithms have been proposed for a variety of scheduling conditions (graph types, granularities or cost models). Jobs can be modeled as map-reduce tasks to exploit maximum parallelization in order to reduce the overall execution time. It is important to device methods exible enough to react to dynamic change of CPU load and memory constraints (data locality of tasks). The main aim of this project is to design a novel algorithm that implements a resource aware scheduling. It is particularly concerned with analyzing dynamically, the data locality of jobs and CPU loads in the nodes of the cluster and scheduling the tasks accordingly. In this project, preemption of tasks is supported. The tasks that are waiting due to non-local data are preempted under failure of threshold times and re-scheduled for better performance.

Multi-Level SVM Based Detection Of DDoS Attacks Using SNMP MIB

Sundaram R (06c100), Sathick Batcha (06c82), Shankaranarayanan (06c76)

In present days, the threat posed by traffic flooding attacks like DDos and Internet worm on private and public computer networks have increased manifold. This has only further increased the importance of intrusion detection and threat classification. Thus it becomes almost mandatory to be able to quickly detect such threats, classify them and neutralize them before they can seriously hamper system performance. However most modern techniques of intrusion detection and classification depend on packet based data to be able to detect attacks and classify them. This has the disadvantage of late detection and high system burden to cope with high speed traffic. 22 MIB variables from SNMP gathered statistical data are used. A feature selection mechanism is used for effective variable selection. Then an update time prediction algorithm is used for effective data gathering. After that a hierarchical SVM structure is used for attack detection and classification. In the level-1 normal and attack traffic are separated and in the level-2, the attack is classified. With data gathered by generating real time TCP Sync, UDP, ICMP, Smurf flooding attacks, the accuracy, precision , recall and F -measure have been measured. Our approach proves to be light on system burden, provides an attack detection accuracy of 95.67% with false alarm rates as low as 1.8%.