Scalable Constraint-based Virtual Data Center Allocation
Authors: Sam Bayless, Nodir Kodirov, Ivan Beschastnikh, Holger H. Hoos, Alan J. Hu
IJCAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 5 Evaluation We now present results from an extensive empirical evaluation demonstrating that our approach offers substantial advantages compared to state-of-the-art methods for VDC allocation. Specifically, we compare the performance of NETSOLVER to that of Second Net s VDCAlloc [Guo et al., 2010] a seminal, sound VDC allocation algorithm with end-to-end bandwidth guarantees and the Z3-based abstraction-refinement procedure from [Yuan et al., 2013], which resembles our approach in that it makes use of an SMT solver. |
| Researcher Affiliation | Academia | Sam Bayless Nodir Kodirov Ivan Beschastnikh Holger H. Hoos University of British Columbia, Canada Universiteit Leiden, The Netherlands {sbayless, knodir, bestchai, hoos, ajh}@cs.ubc.ca |
| Pseudocode | No | The paper describes methodological steps and problem formulations in prose and uses diagrams to illustrate concepts, but it does not contain any structured pseudocode or algorithm blocks with clear labels like 'Algorithm' or 'Pseudocode'. |
| Open Source Code | No | The paper thanks the authors of [Guo et al., 2010] and [Yuan et al., 2013] for making their implementations available, but it does not provide an explicit statement or a link indicating that the source code for NETSOLVER or the methodology described in this paper is publicly available. |
| Open Datasets | Yes | Our first experiment reproduces and extends an experiment from [Yuan et al., 2013], in which a series of identical VDCs is allocated one-by-one to tree-structured data centers... The Second Net benchmark instances are extremely large... The second experiment we conducted is a direct comparison against the original Second Net implementation (which we also used for all comparisons reported later). |
| Dataset Splits | No | The paper describes generating sets of VDCs and repeatedly allocating instances until saturation to evaluate the algorithm's performance, but it does not provide specific training/validation/test dataset splits commonly used in machine learning for model reproduction. |
| Hardware Specification | Yes | Except where noted, all experiments were run on a machine with a 2.66GHz (12MB L3 cache) Intel x5650 processor, running Ubuntu 12.04 and limited to 16GB RAM. |
| Software Dependencies | No | The paper mentions using 'MONOSAT' and 'Z3', and notes the operating system 'Ubuntu 12.04', but it does not provide specific version numbers for these or any other software dependencies (e.g., libraries, solvers) required for exact replication. |
| Experiment Setup | Yes | In each experiment, the algorithms repeatedly allocate VDCs to the data center until they are unable to make further allocations (or until a 1 CPU hour timeout is reached). For these experiments, we generated sets of 10 VDCs each of several sizes (6, 9, 12 and 15 VMs)... We considered 5 different VM sizes, ranging from 1 CPU and 1GB RAM, to 8 CPUs and 16GB of RAM; for our experiments, the slave VMs were selected at random from this set, with the master VM also randomized but always at least as large as the largest slave VM. The Hadoop master has tree connectivity with all slaves, with either 1 or 2 Gbps links connecting the master to each slave. |