On the Cluster Admission Problem for Cloud Computing

Authors: Ludwig Dierks, Ian A. Kash, Sven Seuken

JAIR 2021 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Via extensive simulations grounded in a trace from Microsoft Azure, we show that our admission policies lead to a substantial improvement over the simple threshold policy. We then show that substantial further gains are possible if high-quality information is available about arriving workloads. Based on this, we propose an information elicitation approach to incentivize users to provide this information and simulate its effects.
Researcher Affiliation Academia Ludwig Dierks EMAIL University of Zurich Ian A. Kash EMAIL University of Illinois at Chicago Sven Seuken EMAIL University of Zurich
Pseudocode No The paper describes mathematical formulations and propositions but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any explicit statements about releasing code or links to source code repositories.
Open Datasets Yes In this section, we evaluate the performance of our admission policies using a model fitted to the real-world data trace of Cortez et al. (2017). Cortez et al. (2017) published a data trace consisting of all deployments that populated a Microsoft Azure datacenter in one month.
Dataset Splits No The paper describes a simulation setup based on fitting a model to a data trace, but it does not specify any training/test/validation splits for machine learning experiments. It discusses simulating clusters over a 3-year period.
Hardware Specification Yes With this approach, simulating a single cluster over its complete lifetime of three years takes between 0.5 and 4 core hours on an Intel Xeon Processor E5-2680 v4.
Software Dependencies No The paper discusses mathematical models and distributions (e.g., Gamma distributions, Poisson distributions) but does not specify any software names with version numbers used for implementation or analysis.
Experiment Setup Yes We simulate clusters with capacity c = 20, 000 for a 3-year period with all three policies. An average of 1 new deployment per hour arrives according to a Poisson process. The parameters of each arriving deployment are drawn from the fitted distributions presented in Table 1. Furthermore, we set an SLA of 0.01%. We tune the threshold for each policy via binary search, performing 500 runs with each candidate policy parameter, resulting in an overall runtime on the order of 20, 000 core hours per policy.