No Agent Left Behind: Dynamic Fair Division of Multiple Resources

Authors: I. Kash, A. D. Procaccia, N. Shah

JAIR 2014 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Indeed, in Section 6, we test our mechanisms on real data obtained from a trace of workloads on a Google cluster, and obtain encouraging results. Our next goal is to analyze the performance of both mechanisms on real data, for two natural objectives: the sum of dominant shares (the maxsum objective) and the minimum dominant share (the maxmin objective) of the agents present in the system.
Researcher Affiliation Collaboration Ian Kash EMAIL Microsoft Research Cambridge, UK Ariel D. Procaccia EMAIL Carnegie Mellon University, USA Nisarg Shah EMAIL Carnegie Mellon University, USA
Pseudocode Yes ALGORITHM 1: DYNAMIC DRF ALGORITHM 2: CAUTIOUS LP
Open Source Code No The paper does not contain any explicit statements or links indicating that the source code for the described methodology is publicly available.
Open Datasets Yes As our data we use traces of real workloads on a Google compute cell, from a 7 hour period in 2011 (Hellerstein, 2010). The workload consists of tasks, where each task ran on a single machine, and consumed memory and one or more cores; the demands fit our model with two resources. (Hellerstein, J. L. (2010). Google cluster data. Google research blog. Posted at http://googleresearch.blogspot.com/2010/01/google-cluster-data.html.)
Dataset Splits No The paper mentions sampling strategies and averaging over simulations but does not provide specific training/test/validation dataset splits or cross-validation setups. It states: "For various values of n, we sampled n random positive demand vectors from the traces and analyzed the value of the two objective functions under DYNAMIC DRF and CAUTIOUS LP along with the corresponding lower and upper bounds. We averaged over 1000 such simulations to obtain data points."
Hardware Specification No The paper does not provide specific details about the hardware used to run the experiments.
Software Dependencies No The paper does not list specific software components with version numbers used in the experiments.
Experiment Setup Yes For various values of n, we sampled n random positive demand vectors from the traces and analyzed the value of the two objective functions under DYNAMIC DRF and CAUTIOUS LP along with the corresponding lower and upper bounds. We averaged over 1000 such simulations to obtain data points. Figures 4(a) and 4(b) show the maxsum values achieved by the different mechanisms, for 20 agents and 100 agents respectively. Figures 4(c) and 4(d) show the maxmin values achieved for 20 agents and 100 agents, respectively.