Joint Virtual Machine Embedding and Wireless Data Center Topology Management


Butun B., ONUR E.

17th International Conference on Network and Service Management, CNSM 2021, Virtual, Online, Türkiye, 25 - 29 Ekim 2021, ss.63-69 identifier identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.23919/cnsm52442.2021.9615558
  • Basıldığı Şehir: Virtual, Online
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.63-69
  • Anahtar Kelimeler: Wireless data center network, 60 GHz, Wireless communication, EFFICIENT
  • Orta Doğu Teknik Üniversitesi Adresli: Evet

Özet

© 2021 IFIP.Data Center Networks (DCNs) have become critical infrastructures with the enormous increase in the amount of data generated. This increase introduces some problems in DCNs such as high cabling complexity, high-level power consumption, low space utilization, and oversubscription. Wireless DCNs (WDCNs) address these problems by decreasing cabling complexity and power consumption and increasing space utilization thanks to their incorporation of wireless communication technologies. Moreover, with its highly flexible structure, WDCNs efficiently utilize the bandwidth and provide a promising solution to the oversubscription problem. However, WDCNs pose some challenges such as meeting dynamic traffic demands and increasing throughput in the data center. In this work, we propose a WDCN with dynamic topology, and we aim to handle the dynamic traffic demand of virtual machines (i.e., services) and maximize throughput by efficiently embedding virtual machines into physical machines (i.e., servers) and deploying wireless transceivers on top of the racks to establish dynamic wireless communication using unlicensed 60 GHz bands. To maximize throughput in the WDCN, we create a mixed-integer programming (MIP) problem. Due to the complexity of the problem, we also introduce two heuristics named Heuristic for Wireless Link and Service Deployment (HWSD) and Improved HWSD (I-HWSD). These heuristics significantly improve the time it takes to solve the MIP problem with an optimization solver and approach the optimal solution by about 20%. Unlike top-of-rack (ToR)-to-ToR WDCN, which doesn't consider deploying virtual machines, we increase the throughput we can achieve by deploying virtual machines according to the average traffic demands between them.