I build efficient infrastructure for distributed AI and data-intensive workloads
I am an AI systems and performance engineer and researcher at NVIDIA, focused on building efficient, reliable systems for large-scale AI and data-intensive workloads. My work spans the entire AI infrastructure stack, including high-performance networking, GPU communication, use of hardware accelerators, distributed training and inference, optimizing data intensive applications, with an emphasis on improving performance, resource utilization, and reliability end to end.
I hold a PhD in Computer Science from the Systems Group and also teach data center networking at ETH Zurich.
What I work on
Data center networking
AI workload-aware networking, RDMA, NCCL, Spectrum-X, congestion control, network topologies, and performance analysis for large-scale clusters.
AI infrastructure
Distributed training and inference across GPU and DPU clusters, modeling behavior across compute, memory, storage, and network, and turning measurements into system-level architecture decisions.
Distributed data systems
Distributed data platforms, lakehouse workloads, OLTP and OLAP, streaming, vector data, scheduling, resource balancing, recovery, and production performance.
The common thread is connecting workload behavior to the systems underneath it, then translating performance insights into faster, more predictable, and more efficient distributed AI infrastructure.
Contact
vojislav@dukic.ch
Zurich, Switzerland
Research
My research connects application behavior with distributed execution and network performance. My current work is focused on distributed AI infrastructure, while my earlier research explored cloud systems, data-center networking, and workload-aware performance. Selected work is summarized below.
PhD thesis [ pdf ]
Vojislav Dukic
Beyond the Mega-Data Center: Networking Multi-Data Center RegionsSIGCOMM 20 [ pdf ] [ video ]
Vojislav Dukic, Ginni Khanna, Christos Gkantsidis, Thomas Karagiannis, Francesca Parmigiani, Ankit Singla, Mark Filer, Jeffrey L. Cox, Anna Ptasznik, Nick Harland, Winston Saunders, Christian Belady
Is advance knowledge of flow sizes a plausible assumption?NSDI 19 [ pdf ]
Vojislav Dukic, Sangeetha Abdu Jyothi, Bojan Karlas, Muhsen Owaida, Ce Zhang, Ankit Singla