Research projects

GENI VIOLIN

GENI-VIOLIN

 GENI-VIOLIN's goal is to provide snapshotting facility for entire GENI slices. We leverage the VIOLIN project's VNSnap functionality developed by Prof. Dongyan Xu's group at the Purdue University. The snapshotting facility for the entire slices can be used for fault tolerance, debugging and slice management. The main challenge here is to achieve this with minimal application disruption and no changes to applications or operating system. A successful demonstration of nation-wide snapshot/restore is shown in GEC9 plenary session. More >>>
LiteGreen

LiteGreen

 Great amount of research effort is spent on saving energy in enterprise data centers due to the rise in costs from cooling. There is good amount of energy that is wasted in enterprise desktops as well. LiteGreen is an effort to save desktop energy using virtual machine migration. In LiteGreen, the desktop always runs in a VM, and if the desktop is idle, it will be migrated to a server. The idleness can be defined in various ways, and as soon as the VM is active, a remote desktop session is initiated so that user can start using the desktop immediately. A VM migration is also initiated in the background to bring the VM back to local machine. While the VM is on the server, the VM's memory is reduced to consolidate many idle VMs together. LiteGreen won the best paper award at the USENIX ATC'10. More >>>

Adaptive Data Centers

 Today's data centers built by companies like Google, HP, Amazon and Yahoo host large enterprise applications and incur significant cost in terms of maintenance, power and software. Due to their inherent complexity, there is a lot of interest to consolidate into smaller manageable systems with similar performance. We have been working to develop techniques using classical control theory to adaptive adjust data centers to provide good performance and utilization at the same time. More >>>

Grid File Systems

 In a grid, data is stored in geographically dispersed virtual organizations with varying administrative policies and structure. Current middleware provide basic data management services including data access, transfer and simple replica management. Grid applications often require much more sophisticated and flexible mechanisms for manipulating the data. Some of the requirements include logical hierarchical name space, POSIX interfaces, automatic replica management and latency management. I am developing Gvu, a view-oriented framework, that builds on top of existing grid file systems and supports application or user specific logical views that can be formed using sophisticated queries. More >>>
GridOS Various middleware like Globus, Legion and UNICORE provide software infrastructure for developing applications for the grid. However, operating system support for grid computing is minimal or non-existent. GridOS consists of operating system services that provide mechanisms for high performance I/O, communication, resource management and process management. These services are designed to be modular, policy-neutral, consistent and clean. GridOS is developed as a layered architecture so that higher level modules can be developed on top of core modules. I developed high performance FTP modules and experiments indicate that GridOS outperforms the standard FTP and GridFTP applications. More >>>
ocean (Open Computation Exchange and Auctioning Network) provides software infrastructure to support automated commercial buying and selling of dynamic distributed computing resources over the Internet. OCEAN aims to build a marketplace where resources like CPU time, associated memory usage and network bandwidth are the traded commodities. I investigated the following important problems while working in OCEAN.
  1. How do we provide a market for buying and selling resources?
  2. How do we match the resource requests from buyers with potential sellers?
I developed a self-evolving scalable matching network for finding resources quickly. More >>>
Sphinx is a scheduling middleware for scheduling data intensive application on a dynamically changing grid. It is a project developed under the auspices of the GriPhyN (Grid Physics Network) project. One of the important aspects of Sphinx is the efficient management of data for overall optimal work-flow. The dynamic nature of resources and jobs poses a significant challenge in achieving this. I woked on developing an adaptive data management component(DMC) over existing middleware for achieving efficient data management including optimal transfer of data, replica management and data transfer prediction. More >>>

Miscellaneous