|Email: mgawade at soe.ucsc.edu|
I am a second year graduate student at the University of California, Santa Cruz.
I work in Storage Systems Research Centre (SSRC).
Professor Ethan Miller is my advisor. 1>Performance analysis in distributed
I did my undergradution in Computer Science in India from Vishwakarma Institute of Technology,Pune (VIT) in 2004. I worked in industry for 3 years before coming to graduate school in Fall 2007. Here is my resume.Resume.
My main interests are towards performance management in storage and distributed systems.
1>Performance analysis in distributed systems:Designed and developed a multithreaded framework for monitori ng anomalous nodes in a large scale distributed system.The scalability was achieved by using a decentralized hierarchical architecture. Disk throughput was considered a performance matrix for comparison amongst different nodes. Proximity based outlier detection scheme was used for detecting outlier nodes.
2>Map/Reduce performance optimization:
The performance of a Map/Reduce query depends critically on the skew of the parallel computation in the Reduce operators. Essentially, the slowest Reduce operator determines the speed of the overall computation. This project is about determining methods inspired from Peer-to-Peer systems in order to auto-tune the number of Reduce operators so as to avoid computational skew.
3>De-duplication in log structured file systems:
Studied the feasibility of de-duplication support in a “Log structured file system” in Linux. De-duplication is detected by using chunking based mechanism based on Rabin Fingerprint algorithm.
4>Role of Buffer caches in quality of service guarantees:
Client buffer cache and server buffer cache exhibit different behavior for quality of service guarantees. Survey different aspects of buffer caches, which matter in guaranteeing the quality of service in a distributed environment. Different aspects that affect performance are buffer allocation, pre-fetching, cache partitioning, hit rate, slack management, type of workload, block replacement policy, collaboration between client and server cache.
5>Comparison of I/O schedulers:
Compared two existing disk schedulers named Argon and Fahrrad on the basis of throughput efficiency. Argon is implemented as a prototype Linux module using existing Fahrrad framework for a fair comparison. Study of sequential, semi-sequential, random access patterns issued by workload is done. Main aim is to study how the two schedulers behave under mixed workload conditions and how changes in the workload affect the throughput guarantees reserved by applications with schedulers.