Availability-aware Mobile Edge Application Placement in 5G Networks He Zhu † , Changcheng Huang † † Department of Systems and Computer Engineering Carleton University, Ottawa, ON K1S 5B6, Canada {hzhu, huang}@carleton.ca Abstract—Mobile edge computing (MEC) literally pushes cloud computing from remote datacenters to the life radius of end users. By leveraging the widely adopted ETSI network function virtualization (NFV) architecture, MEC provisions elastic and resilient mobile edge applications with proximity. Typical MEC virtualization infrastructure allows configurable placement policy to deploy mobile edge applications as virtual machines (VMs): affinity can be used to put VMs on the same host for inter-VM networking performance, while anti-affinity is to separate VMs for high availability. In this paper, we propose a novel model to track the availability and cost impact from placement policy changes of the mobile edge applications. We formulate our model as a stochastic programming problem. To minimize complexity challenge, we also propose a heuristic algorithm. With our model, the unit resource cost increases when there are less resources left on a host. Applying affinity would take up more resources of the host but saves network bandwidth cost because of co-location. When enforcing anti-affinity, experimental results show increases of both availability and inter-host network bandwidth cost. For applications with different resource requirements, our model is able to find their sweet points with the consideration of both resource cost and application availability, which is vital in a less robust MEC cloud environment. Keywords—Mobile Edge Computing, 5G, Placement Policy, Stochastic Optimization, Cloud Computing. I. I NTRODUCTION The emerging edge computing is bringing all benefits of cloud computing to mobile devices plus proximity [1], to deliver highly-responsive cloud services at the network edge. As a key technology towards 5G, the mobile edge computing (MEC) architecture proposed by ETSI leverages the widely adopted frameworks of Network Function Vir- tualization (NFV) [2]. Elastic mobile edge applications are deployed close to the user equipment (UE) with low latency. Both UE application providers and telecommunication service providers (TSPs) can take advantage of MEC to reduce costs and to adjust services with agility based on fast-changing user demands. A mobile edge application consists of one or more collab- orating virtual machines (VMs). It is of paramount impor- tance to maintain the high availability of MEC applications. Compared to centralized datacenters used by public cloud, MEC hosts are heterogeneous with varying computing, storage and networking capabilities [3]. Smaller scale private cloud servers can be deployed near their designated groups of Fig. 1. A mobile edge application deployment with placement rules. There are five VMs deployed in three groups with each group placed on a separate host. A minimum of three VMs are required for the application. The placement will ensure the application is in service if one host is down. users as MEC hosts, the characteristics of which lead to the following indications: (i) A single MEC server deployment is less powerful because it serves a smaller group of users. It is unlikely to merit its own security guard or have the same level of redundancy as a larger facility [4]. (ii) The offloading nature of MEC brings higher system complexity which may jeopardize availability [5]. These facts conclude that hosts in MEC are less reliable. When mobile edge applications run in MEC servers, they must be protected from host failure. To maximize the availability while maintaining costs and la- tencies at acceptable levels, placement rules, which determine on which computing host each VM is deployed, often come into play to tune the performance and security of a mobile edge application [6]. In practice, placement rules mainly refer to the affinity and anti-affinity rule [7]. The affinity rule helps reduce communication costs between VMs serving the same mobile edge application: VMs on the same host connect to each other using virtual networks and require no physical networking infrastructure. Same-host network traffic essentially takes up computing capabilities of the host and has much better performance than physical networks. This becomes handy especially when frequent inter- VM communications are needed. An obvious down side of the affinity rule is putting all eggs in one basket. If the host is