Lloyd Brown

EECS Department, University of California, Berkeley

Technical Report No. UCB/EECS-2025-219

December 19, 2025

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2025/EECS-2025-219.pdf

This dissertation presents two frameworks that provide solutions to the problems that arise when contention occurs that better align with the incentives of the underlying system than the state of the art solution. Contention is an important part of large-scale systems. In large-scale systems it becomes untenable to provision for the peak usage due to the sheer scale of resource needs. As a result these systems must handle periods of overuse by allocating the available resources based on the overarching goals of the system. In both of these frameworks we: discuss the nature of the contention that happens in the problem area, outline the desirable properties of an ideal resource allocation scheme, find that existing allocation schemes fall short with respect to those properties, and propose an alternative solution that is better aligned with those properties. In the first framework, RCS, we focus on the wide-area Internet and propose a new set of guidelines for bandwidth allocations under congestion. In the second framework, Rose, we focus on cluster management systems and propose a novel scheduler and checkpointing mechanism to reduce cycles lost due to eviction during periods of resource oversubscription.

Advisors: Scott Shenker

---

BibTeX citation:

@phdthesis{Brown:EECS-2025-219,
    Author= {Brown, Lloyd},
    Title= {Managing Contention in Large-scale Systems},
    School= {EECS Department, University of California, Berkeley},
    Year= {2025},
    Month= {Dec},
    Url= {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2025/EECS-2025-219.html},
    Number= {UCB/EECS-2025-219},
    Abstract= {This dissertation presents two frameworks that provide solutions to the problems that arise when contention occurs that better align with the incentives of the underlying system than the state of the art solution. Contention is an important part of large-scale systems. In large-scale systems it becomes untenable to provision for the peak usage due to the sheer scale of resource needs. As a result these systems must handle periods of overuse by allocating the available resources based on the overarching goals of the system. In both of these frameworks we: discuss the nature of the contention
that happens in the problem area, outline the desirable properties of an ideal resource allocation scheme, find that existing allocation schemes fall short with respect to those properties, and propose an alternative solution that is better aligned with
those properties. In the first framework, RCS, we focus on the wide-area Internet and propose a new set of guidelines for bandwidth allocations under congestion. In the second framework, Rose, we focus on cluster management systems and propose
a novel scheduler and checkpointing mechanism to reduce cycles lost due to eviction during periods of resource oversubscription.},
}

EndNote citation:

%0 Thesis
%A Brown, Lloyd 
%T Managing Contention in Large-scale Systems
%I EECS Department, University of California, Berkeley
%D 2025
%8 December 19
%@ UCB/EECS-2025-219
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2025/EECS-2025-219.html
%F Brown:EECS-2025-219