A Mean-Field Game Approach to Cloud Resource Management with Function Approximation
Authors: Weichao Mao, Haoran Qiu, Chen Wang, Hubertus Franke, Zbigniew Kalbarczyk, Ravishankar Iyer, Tamer Basar
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We further implement our algorithm using both linear and neural-network function approximations, and evaluate our solution on an open-source serverless platform, Open Whisk, with real-world workloads from production traces. Experimental results demonstrate that our approach is scalable to a large number of users and significantly outperforms various baselines in terms of function latency and resource utilization efficiency. |
| Researcher Affiliation | Collaboration | Weichao Mao University of Illinois Urbana-Champaign weichao2@illinois.edu; Haoran Qiu University of Illinois Urbana-Champaign haoranq4@illinois.edu; Chen Wang IBM Research chen.wang1@ibm.com; Hubertus Franke IBM Research frankeh@us.ibm.com; Zbigniew Kalbarczyk University of Illinois Urbana-Champaign kalbarcz@illinois.edu; Ravishankar K. Iyer University of Illinois Urbana-Champaign rkiyer@illinois.edu; Tamer Ba sar University of Illinois Urbana-Champaign basar1@illinois.edu |
| Pseudocode | Yes | Algorithm 1: Natural Actor-Critic for MFGs with Linear Function Approximation |
| Open Source Code | No | The checklist states 'Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes]' but no specific URL or instruction on how to access the code is provided within the paper. |
| Open Datasets | Yes | To drive the benchmarks, we sample and replay the function invocations from Azure function traces [61]. |
| Dataset Splits | No | The paper does not explicitly provide training/test/validation dataset splits, specific percentages, or absolute sample counts for each split. |
| Hardware Specification | No | The paper mentions deploying on "IBM Cloud with 22 VMs" but does not specify any particular hardware details such as GPU models, CPU models, or memory configurations for these virtual machines. |
| Software Dependencies | No | The paper does not provide a reproducible description of ancillary software, such as specific library names with version numbers (e.g., Python, PyTorch, CUDA versions). |
| Experiment Setup | Yes | We use a production-grade open-source serverless platform, Open Whisk [22], and deploy it on IBM Cloud with 22 VMs... We implement and evaluate two variants of our method: one exploits linear function approximation for both the actor and the critic ( NAC-Linear for short), and the other one leverages a two-layer fully-connected neural network as function approximation (i.e., NAC-NN )... For training, we create 50 functions3 on Open Whisk, each of which is randomly selected from the function benchmarks... we consider a reward function as rt = α QP(t)+(1 α)/2 (RUcpu(t)+RUmem(t))+penalty, where QP(t) and RU(t) are the Qo S preservation ratio and resource utilization at time t, and penalty is set to -1 (and 0 otherwise) for illegal or undesired actions (e.g., dangling decisions). |