Research Areas

My research interests lie at the intersection of network protocols, transport-layer reliability, distributed systems, and space networking. More broadly, I study how communication and stream-processing systems can be made more robust, adaptable, and efficient under dynamic network conditions, heterogeneous resources, long disruptions, and large-scale data movement.

I am currently focusing on the following research directions:

  • Recursive architectures for reliable large-scale data transfer: through the Multi-Level Error Detection (MLED) framework, I study how configurable recursive layers and in-network resources can reduce undetected error probability, localize recovery, and improve goodput in large-scale file transfers.
  • QUIC behavior over Internet-scale WAN paths: I investigate how path degradation, PTO behavior, 4-tuple variation, and port migration affect QUIC performance, latency, recovery, and connection stability under dynamic network conditions.
  • Adaptive execution in hybrid edge–cloud systems: I explore how datacenter-centric stream-processing engines such as Apache Flink can support heterogeneous edge–cloud deployments through automatic query rewriting, adaptive routing, and dynamic operator placement while preserving event-time processing and fault tolerance.
  • Recursive space networking architectures: I study how scoped communication domains, service intent, durable service semantics, and contact-aware forwarding can be combined into a unified architecture for challenged space environments such as lunar, cislunar, and interplanetary networks.

Research Projects

Multi-Level Error Detection (MLED) Framework

View project details →

The Multi-Level Error Detection (MLED) framework is a configurable recursive architecture for large-scale file transfer that leverages in-network resources to reduce the probability of undetected errors and localize recovery within the network. It is designed for scientific and data-intensive settings where even a single undetected error can invalidate a valuable dataset.

MLED generalizes the traditional two-layer error-detection approach by introducing additional levels with configurable policies over different scopes. It has been mathematically analyzed and experimentally validated on the FABRIC testbed, where it detected and corrected adversarial errors inside the network, achieved a 100% goodput gain over the traditional approach under non-zero error rates, and reached over 800 Mbps goodput on a single connection.

You can access the experimental setup, software artifacts, and implementation resources here:

Heterogeneity-Aware Adaptive Placement in Hybrid Edge–Cloud Systems

View project details →

This project explores how datacenter-centric stream-processing engines such as Apache Flink can be extended to support hybrid edge–cloud analytics without requiring users to rewrite their applications. The work focuses on preserving built-in capabilities such as fault tolerance, event-time processing, and ecosystem compatibility while enabling deployment across heterogeneous edge and cloud resources.

The framework uses automatic query rewriting, adaptive routing, and dynamic operator placement to shift lightweight stateless operators between edge devices and the cloud under changing bandwidth, workload, and compute conditions. It has been evaluated on a physical Raspberry Pi testbed and on a large-scale 100-VM deployment across multiple real-world streaming applications.

ReSpaN: Recursive Space Networking Architecture

View project details →

ReSpaN is a proposed recursive space-networking architecture for challenged environments such as lunar, cislunar, and interplanetary communication. It builds networking from scoped Distributed IPC Facilities (DIFs) rather than a rigid fixed stack, allowing each scope to apply policies appropriate to its delay, disruption, trust, and resource constraints.

The architecture combines service intent, a three-plane structure spanning Control, Service, and Data planes, and label-based fast forwarding to support durable object delivery, contact-aware planning, and efficient in-contact transfer. It is designed to contain churn locally while preserving end-to-end communication semantics across heterogeneous space environments.