Salini S Kowsalya

 

 

Projects

 

1. Multilevel I/O tagging in a Virtualized Environment

 

Abstract: There is a huge semantic gap between file system and storage in servicing an I/O request. In a virtual environment the gap increases even more, as the hypervisor’s I/O scheduler is not aware of the semantics of the I/O request that the application issues. To address this limitation and to retain the semantics of the I/O requests in a virtualized environment, we tag each I/O request from the guest application all the way down to the hypervisor’s I/O scheduler. By segregating I/O requests, the hypervisor can give differentiated I/O services to different applications both within and across the guest virtual machines. In this paper, we present the design and implementation of such a system on Kernel Based Virtual Machine. We evaluated our system on hard disks and solid state disks for different workloads. Our results show that preferential services are well achieved for applications running within and across guest virtual machines.
Paper   Presentation Slides

 

2. Website Fingerprinting using Traffic Analysis

 

Abstract: Website fingerprinting is the act of recognizing web traffic through surveillance despite the use of encryption or anonymizing software. The overall idea is to leverage the fact that many web sites have specific request patterns, response byte counts, and other similar coarse features that are known beforehand. This information can be used to recognize and classify different website traffic despite attempts at encryption or tunneling. In this paper, I experimentally prove this concept. I extract coarse features of network traffic such as bandwidth, packet length, inter-arrival time, average packets sent/received per second of five popular websites namely, gmail, skype, facebook, gmail video chat, and youtube traffic and train the system. After sufficient training, I run the system on randomly collected traffic samples and calculated the accuracy of classification. During my experiments, I also observed that each website has different traffic profiles with respect to different browsers. Hence one cannot achieve accuracy in classification of traffic samples by training the system using a set of features obtained from traffic profiles collected using one browser and applying the same set of features on the sample traffic profiles collected using a different browser even for the same website.
Paper   Presentation Slides

 

3. Performance Analysis of Aerie under mail server workloads

 

Abstract: Inspite of the rapid advancements in storage technology, the fundamental architecture of storage in operating systtem remains fixed. Applications invoke the kernel to store and retrieve data and kernel invokes the file system. Recent work suggests a new class of memory called storage class memory(SCM) which blurs out the distinction between fast, expensive and volatile memory, and slow, cheap and non-volatile storage. The file system for such a class of memory does not need kernel interaction. Applications can modify the file system and optimize it according to their needs. One such flexible file system architecture is provided by Aerie. It exposes storage class memory to user mode programs which allows dirrect access to memory from user mode. Aerie evaluates the application level performance on three Filebench profiles: file server, web server and web proxy. We are interested in measuring the performance of Aerie for mail server profiles. Mail servers usually deal with a large number of small files which are in constant flux. It has frequent reads/writes, create/delete and appends. In this paper, we adopt two simple benchmarks which mimic the behavior of a mail server and alter them to use the file system APIs exposed by Aerie. We compare the performance of Aerie with tmpfs and ext3 file system (with and without caching) on linux 3.2.2.
Paper