Title: End-to-End Support for In-Network Acceleration of Dynamically Chained Parametric Tasks

Rafael Oliveira  

Ph.D. Student in Computer Science  

School of Computer Science  

College of Computing  

Georgia Institute of Technology  

Date: Tuesday, September 30th, 2025  

Time: 3:00pm - 4:30pm ET  

Location: KACB 3100, or Zoom

Committee:  

Dr. Ada Gavrilovska  (advisor) – College of Computing, Georgia Institute of Technology  

Dr. Calton Pu – College of Computing, Georgia Institute of Technology  

Dr. Santosh Pande – College of Computing, Georgia Institute of Technology  

Dr. Alex Daglis – College of Computing, Georgia Institute of Technology  

Dr. Hadi Esmaeilzadeh – Computer Science and Engineering, UCSD  

Abstract: Modern cloud infrastructure faces a fundamental inefficiency: while modern software scales through shared resources—from dynamically linked libraries to containerized microservices—SmartNICs allocate dedicated hardware accelerators per tenant, evoking the pre-DLL era where common functions were duplicated in every program. This wastes precious hardware resources in multi-tenant environments where applications often need the same computations (AES encryption, Kalman filters) with different parameters.

This thesis introduces parametric Compute Unit (CU) sharing—enabling multiple tenants to share hardware accelerators just as software shares libraries. The key challenge is managing parameters at line rate: unlike software's flexible memory, hardware requires deterministic, cycle-accurate parameter delivery across dynamically composed processing chains.

We present three systems that collectively solve this challenge: Vazado enables dynamic chaining of CUs with incompatible signatures through hardware-compiler co-design, achieving 13.19× energy efficiency over 48-core CPUs while using just 10% of FPGA resources. ErdTree introduces hierarchical topic trees that reduce memory footprints by 52% through automatic parameter deduplication and hardware-friendly FIFO caching. COMPREX accelerates NIC-CPU transfers 7× through domain-specific compression.

Together, these systems demonstrate that efficient multi-tenant acceleration is achievable without sacrificing programmability—enabling SmartNICs to finally deliver on their promise of scalable, shared acceleration for cloud infrastructure.