My research focus is in experimental systems: operating system topics in multimedia, storage, security, networks and sensor systems. Some of the research areas that I am actively involved with include:
Laptops are resource rich and are gradually replacing desktops as the primary computing platform for many users. Laptops are mobile with unreliable network connectivity. Traditionally laptops were dependent on the resources provided by established storage infrastructure. They mostly operated alone with limited sharing of their storage resources with other laptops. We are investigating the viability of federating the storage resources available on laptops. The research is motivated by a desire to allow students in an university setting to share their individual laptop storage in order to achieve a collaborative storage for the instructors, other students who are registered in the same courses as well as other social group members. The amount of federated resources are expected to be on the order of hundreds of TBs for our university.
We are currently investigaing the viability of using laptops for shared storage by analyzing the mobility behavior of a large number of laptop users in a university setting. The preliminary analysis shows that the availability characteristics were worse than was observed in a corporate desktop setting. However, the temporal consistency wherein laptops were either
simultaneously available or unavailable were higher. Laptop users also appear to exert control over sharing durations; presumably to conserve local battery and other resources. These observations preclude synchronous storage such as
Farsite; asynchronous storage such as Bayou appear more relevant. However, the availability characteristics affect the rates of replica divergence and amounts of roll-back required in Bayou. We are researching a system where laptops act as stowaways and store updates for unrelated laptops. The stowaway storage is the currency required for laptops to participate in the federation. The system maintains consistency at the granularity of files. Updates are propagated via the stowaway storage using epidemic algorithms. Updates are applied conservatively in order to avoid unnecessary roll-backs. We will unify the namespace across all simultaneously online laptops. We are also developing the notion of temporal and spatial correlation sets that will allow the laptops to use the contents of the stowaway storage and provide some consistency guarantees.
This research was motivated by the mismatch between the global resource requirements of peer computing systems and the local operating system resource management mechanisms. Peer computing systems such as ad hoc networks and peer-to-peer (p2p) sharing systems federate the spare resources available among a set of independent and cooperating peers. Even though there is a wealth of research on federation mechanisms, the precise resource allocation responsibilities of the peer to the federation is undefined. The operating system on the peer allocates local resources to all scheduleable entities; even low priority objects are eventually assigned the requested resources. Operating systems are unaware of peer resource allocation assumptions and hence peer resource requests are treated as equal or in the worst case as more important that of local requests. In practice, this means that resource constrained peers continue to provide resources to other peers while jeopardizing resource availability for local users. Traditional operating systems are ill equipped to manage spare resources for peer computing systems. Operating Systems should be explicitly aware of resources consumed by peers and treat peers as second class citizens. Under resource constrained environments, peer resource requests should be intentionally unsatisfied even though there are no other competing local resource requests. Once we develop the resource management mechanisms, we will also revisit the federation mechanisms and make them cooperate with nodes that are aware of peer resource requests.
Traditional scalable, reliable, secure and managed storage is expensive. Sensor networking and peer-to-peer application scenarios operate without any trusted infrastructure and can tolerate weaker guarantees. We are investigating aa fully decentralized security infrastructure that is built over distributed commodity storage. Specifically, the system provides data availability, access control, and versioning with tunable levels of certainty. The stringency of traditional security guarantees for these operations are minimally relaxed. In exchange the protocols are resilient and can continue to operate correctly despite a fraction of the storage nodes being disconnected or acting maliciously. The proposed research will be experimentally validated using a multi-petabyte distributed storage system being built using a combination of university maintained desktops and custom storage units.
We are interested in the implications of wireless services such as AppleTV, Slingbox as well as wireless game units such as Sony PSP, Nintendo DS etc. We have a bunch of these devices and are looking for someone who are good at playing multiplayer video games using these units.
P2p technologies promise decentralized mechanisms to not only access popular objects (that are replicated everywhere), but also locate peers that distribute rare and specialized objects. We envision a p2p system wherein the participating peers operate independently and periodically form a community to share objects. New participating peers frequently enter the system; existing peers leave the system while existing links can degrade in performance. A given peer's view of the network can frequently become obsolete. Stale list of peers can force the system to repeatedly choose bad links as well maintain unncessarily redundant topologies. If care is not taken to maintain the p2p topology, there is also a danger of a p2p application partition.
Such access patterns map naturally into unstructured p2p networks. The users want to identify the closest neighbors that share the particular object of interest. The definition of close depends on the particular application; typical examples include peers with the shortest IP route, highest throughput bandwidth, geographic distance and the most energy to spare for sharing objects etc. Once these neighboring peers are identified, the p2p application can leverage the underlying network mechanisms to search and request the objects. As resources are depleted, the system will reconfigure itself to maintain the topology. We are exploring some of these research challenges.
In this research, we are concerned with the key challenge of developing techniques that transition the system components to lower power consuming states in order to achieve overall end-to-end energy saving. We utilize techniques to exploit the various power states of the mobile devices to achieve overall energy savings. We use application level cooperative scheduling of network resources to reduce the time spent in high power consuming states. We explore the practical implications of a cooperative scheduling mechanism that reduces the amount of time spent in high energy consuming idle states to achieve energy consumption reduction.
This work investigates the web access patterns of mobile devices. We will instrucment a typical web browser, Mozilla, to collect the usage patterns of a target user group. Generated data will be analyzed for such aspects as actual web objects accessed as well as amount of data actually viewed on the screen. This work serves as a starting point for developers of lazy data transmission policies as well as efficient prefetching and caching.
The goal of this research effort is to customize multimedia objects so that they are appropriate in more scenarios than they are in the original form. We utilize transcoding as the enabling technology. Transcoding is a transformation that is used to convert a multimedia object from one form to another (frequently trading off object fidelity for size). For transcoding to be useful, we need to understand the tradeoff characteristics: the information quality loss, the computational overhead required in computing the transcoding and the potential space benefits.
You can learn more about my research areas from my home page (http://www.cse.nd.edu/~surendar/). As such, I have funding for a few students and am always looking for students to colloborate with.
If you are thinking about applying to Notre Dame, I highly encourage you to apply to our program. We have an active research program focussing on a wide variety of topics. We are a Ph.D. only graduate program and all incoming students are funded through fellowships or assistantships (so, funding should not be your concern in talking to faculty). If you are interested, we may be able to arrange for a visit to ND. If you would like more information about systems related research possibilities, please drop me a note at surendar@nd.edu.
If you are currently a student in Notre Dame (both undergraduate and graduate), drop by anytime to chat about research possiblities. I also strongly encourage you to attend the systems seminar; tuesdays at 3:30 pm in Fitz 384C.