We are the Systems and Theory Group for Intelligent and Adaptive Networks (STyGIANet) in the Computer Science Department at Purdue University, led by Vamsi Addanki. Our research sits at the intersection of networked systems, distributed computing, and theory, where we design, model, and analyze the networks that power today’s most demanding applications. We are especially excited about building the next generation of photonic interconnects for GPU clusters, aiming to break through the bandwidth barriers that limit distributed training and inference at scale.

At STyGIANet, we take a systems-driven yet theory-grounded approach:

• We prototype bold new ideas in datacenter networking, from photonic interconnects for GPU clusters to AI-driven protocols.
• We use theory as a lens to analyze performance, guide design choices, and ensure our systems scale in practice.
• We collaborate with industry partners and across disciplines to translate research into deployments, open-source tools, and real-world impact.

Openings

• Purdue students looking for PhD advisor
  We are currently looking for candidates leaning towards theory and with prior networking background to join our group. Please send an email to Vamsi Addanki.
• Purdue undergraduate students
  Drop by at one of our weekly group meetings to have a chat. Please send an email for the location and time.
• External students
  Please email your CV along with a brief motivation.

News

	NSF Grant for STyGIANet! 01-July-2026 Our group receives funding from NSF for the proposal "Accelerating AI Workloads with Programmable Photonic Interconnects" under the Future CoRe program. The award will support research on adaptive photonic interconnects for distributed training and inference.
	Two papers accepted to SIGCOMM 2026! 01-July-2026 Two papers "Harvest: Adaptive Photonic Switching Schedules for Collective Communication in Scale-up Domains" PDF and "Trivance: Latency-Optimal AllReduce by Shortcutting Multiport Networks" PDF accepted to SIGCOMM. Harvest synthesizes photonic switching schedules in a reconfiguration-delay-aware manner, significantly reducing the completion time of AllReduce and All-to-All operations in distributed training scale-up domains. Technically, Harvest is the first work that incorporates reconfiguration delay into Birkhoff decompositions and topology synthesis with multi-hop routing, leveraging the connection to maximum concurrent flow problem. Trivance, is a novel AllReduce algorithm specifically tailored for Torus networks (e.g., TPUv4), completing within $\log_3(n)$ steps at a significantly lower congestion.
	Paper accepted to HotOptics 2026! 10-June-2026 Our paper "Birkhoff Decompositions and Photonic Interconnects Wait! Don’t Forget the Compute!" PDF accepted to HotOptics 2026, highlights the importance of jointly optimizing network communication and expert computation to maximize communication--computation overlap in Mixture-of-Experts execution. It demonstrates that decomposition strategies, which are fundamental to photonic switching, have a significant impact on overlap efficiency. The paper further shows that strategies that are efficient from a networking perspective do not necessarily lead to efficient dependent computation.
	Leaf joins STyGIANet. Welcome! 16-Dec-2025 Leaf joins the group as a PhD student. He is broadly interested in High-Performance Computing, and networking.
	Ishan joins STyGIANet. Welcome! 16-Dec-2025 Ishan joins the group as a masters student. He is interested in scheduling techniques for LLM inference in multi-GPU clusters.
	Doctoral Dissertation Award! 15-Dec-2025 Vamsi's doctoral dissertation at TU Berlin has been awarded Tiburtius Prize, and honorable mention in Sigmetrics Doctoral Dissertation Award 2025.
	Kriti joins STyGIANet. Welcome! 06-Dec-2025 Kriti joins as a sophomore undergraduate student. She is passionate about LLM pipelines, reasoning models, and networking challenges in AI.
	Sam joins STyGIANet. Welcome! 05-Dec-2025 Samuel joins the group with wide-ranging interests in datacenter networking and AI systems. He will be working closely with foundation models such as Llama.
	Nihar joins STyGIANet. Welcome! 04-Nov-2025 Nihar joins as a Junior-year undergraduate student with expertise in MPI, parallel computing, and collective communication libraries. He will be focusing on cuda kernels and dma latencies.
	New paper accepted to SOSA 2026! 05-Oct-2025 Our paper "The Harmonic Policy for Online Buffer Sharing is (2+ln)-Competitive: A Simple Proof" revisits the best known upper bound for drop-tail buffer sharing (Harmonic algorithm) and presents a much simpler proof for its ln-competitive ratio. PDF
	Nirmal joins STyGIANet. Welcome! 03-Oct-2025 Nirmal joins as a Junior-year undergraduate student with expertise in GPU kernels and distributed inference. He will be focusing on LLM inference in GPU clusters.
	New paper accepted to HotNets 2025! 17-Sept-2025 Our paper "When Light Bends to the Collective Will: A Theory and Vision for Adaptive Photonic Scale-up Domains" envisions fully photonic chip-to-chip communication in GPU clusters, characterizes the performance of collective communication, and opens an optimization opportunity for adaptive photonic interconnects. PDF
	Deeksha joins STyGIANet. Welcome! 25-Aug-2025 Deeksha is a Masters student, specializing in FPGAs and verilog. At STyGIANet, she will be focusing on the systems challenges involved in reconfigurable networks.
	Mahir joins STyGIANet. Welcome! 20-Aug-2025 Mahir is a second year PhD student, co-advised by Prof. Sonia Fahmy. Mahir will focus broadly on datacenter networking and collective communication workloads.
	Vamsi joins Purdue CS as Assistant Professor! 18-Aug-2025 Vamsi joins the Computer Science Department as an Assistant Professor on August 18th, 2025. He will lead the STyGIANet lab to build the next generation intelligent and adaptive networks.

NSF Grant for STyGIANet!

01-July-2026

Our group receives funding from NSF for the proposal "Accelerating AI Workloads with Programmable Photonic Interconnects" under the Future CoRe program. The award will support research on adaptive photonic interconnects for distributed training and inference.

Two papers accepted to SIGCOMM 2026!

01-July-2026

Two papers "Harvest: Adaptive Photonic Switching Schedules for Collective Communication in Scale-up Domains" PDF and "Trivance: Latency-Optimal AllReduce by Shortcutting Multiport Networks" PDF accepted to SIGCOMM. Harvest synthesizes photonic switching schedules in a reconfiguration-delay-aware manner, significantly reducing the completion time of AllReduce and All-to-All operations in distributed training scale-up domains. Technically, Harvest is the first work that incorporates reconfiguration delay into Birkhoff decompositions and topology synthesis with multi-hop routing, leveraging the connection to maximum concurrent flow problem. Trivance, is a novel AllReduce algorithm specifically tailored for Torus networks (e.g., TPUv4), completing within $\log_3(n)$ steps at a significantly lower congestion.

Paper accepted to HotOptics 2026!

10-June-2026

Our paper "Birkhoff Decompositions and Photonic Interconnects Wait! Don’t Forget the Compute!" PDF accepted to HotOptics 2026, highlights the importance of jointly optimizing network communication and expert computation to maximize communication--computation overlap in Mixture-of-Experts execution. It demonstrates that decomposition strategies, which are fundamental to photonic switching, have a significant impact on overlap efficiency. The paper further shows that strategies that are efficient from a networking perspective do not necessarily lead to efficient dependent computation.

Leaf joins STyGIANet. Welcome!

16-Dec-2025

Leaf joins the group as a PhD student. He is broadly interested in High-Performance Computing, and networking.

Ishan joins STyGIANet. Welcome!

16-Dec-2025

Ishan joins the group as a masters student. He is interested in scheduling techniques for LLM inference in multi-GPU clusters.

Doctoral Dissertation Award!

15-Dec-2025

Vamsi's doctoral dissertation at TU Berlin has been awarded Tiburtius Prize, and honorable mention in Sigmetrics Doctoral Dissertation Award 2025.

Kriti joins STyGIANet. Welcome!

06-Dec-2025

Kriti joins as a sophomore undergraduate student. She is passionate about LLM pipelines, reasoning models, and networking challenges in AI.

Sam joins STyGIANet. Welcome!

05-Dec-2025

Samuel joins the group with wide-ranging interests in datacenter networking and AI systems. He will be working closely with foundation models such as Llama.

Nihar joins STyGIANet. Welcome!

04-Nov-2025

Nihar joins as a Junior-year undergraduate student with expertise in MPI, parallel computing, and collective communication libraries. He will be focusing on cuda kernels and dma latencies.

New paper accepted to SOSA 2026!

05-Oct-2025

Our paper "The Harmonic Policy for Online Buffer Sharing is (2+ln)-Competitive: A Simple Proof" revisits the best known upper bound for drop-tail buffer sharing (Harmonic algorithm) and presents a much simpler proof for its ln-competitive ratio. PDF

Nirmal joins STyGIANet. Welcome!

03-Oct-2025

Nirmal joins as a Junior-year undergraduate student with expertise in GPU kernels and distributed inference. He will be focusing on LLM inference in GPU clusters.

New paper accepted to HotNets 2025!

17-Sept-2025

Our paper "When Light Bends to the Collective Will: A Theory and Vision for Adaptive Photonic Scale-up Domains" envisions fully photonic chip-to-chip communication in GPU clusters, characterizes the performance of collective communication, and opens an optimization opportunity for adaptive photonic interconnects. PDF

Deeksha joins STyGIANet. Welcome!

25-Aug-2025

Deeksha is a Masters student, specializing in FPGAs and verilog. At STyGIANet, she will be focusing on the systems challenges involved in reconfigurable networks.

Mahir joins STyGIANet. Welcome!

20-Aug-2025

Mahir is a second year PhD student, co-advised by Prof. Sonia Fahmy. Mahir will focus broadly on datacenter networking and collective communication workloads.

Vamsi joins Purdue CS as Assistant Professor!

18-Aug-2025

Vamsi joins the Computer Science Department as an Assistant Professor on August 18th, 2025. He will lead the STyGIANet lab to build the next generation intelligent and adaptive networks.

Older news can be found here