Energy Efficient Placement of Workloads in Composable Data Center Networks

Abstract

This paper studies the energy efficiency of composable data center (DC) infrastructures over network topologies. Using a mixed integer linear programming (MILP) model, we compare the performance of disaggregation at rackscale and pod-scale over selected electrical, optical and hybrid network topologies relative to a traditional DC. Relative to a podscale DC, the results show that physical disaggregation at rackscale is sufficient for optimal efficiency when the optical network topology is adopted, and resource components are allocated in a suitable manner. The optical network topology also enables optimal energy efficiency in composable DCs. The paper also studies logical disaggregation of traditional DC servers over an optical network topology. Relative to physical disaggregation at rack-scale, logical disaggregation of server resources within each rack enables marginal fall in the total DC power consumption (TDPC) due to improved resource demands placement. Hence, an adaptable composable infrastructure that can support both in memory (access) latency sensitive and insensitive workloads is enabled. We also conduct a study of the adoption of micro-service architecture in both traditional and composable DCs. Our results show that increasing the modularity of workloads improves the energy efficiency in traditional DCs, but disproportionate utilization of DC resources persists. A combination of disaggregation and micro-services achieved up to 23% reduction in the TDPC of the traditional DC by enabling optimal resources utilization and energy efficiencies. Finally, we propose a heuristic for energy efficient placement of workloads in composable DCs which replicates the trends produced by the MILP model formulated in this paper. Index Terms-Composable infrastructures, energy efficient data centers, MILP, micro-services, optical networks, rack-scale data center, software defined infrastructures. I. INTRODUCTIONATA CENTERS are critical infrastructures that provide platforms driving wide adoption of digital technologies. These indispensable infrastructures provide computing resources needed to run public internet-facing applications and private enterprise-critical applications alike in environments that support the requirements of cloud computing and data analytics applications. The requirements of cloud computing and data analytic applications include on-demand resource provisioning, multitenant isolation, parallel computation, and security. Examples of such applications include web services,

Network Topologies for Composable Data Centers

“…Resource disaggregation mitigates the resource stranding problem associated with the server centric DCs to enable greater utilization of resource components in cloud DCs [17] and fog DCs [19]. This is achieved by physically and/or logically separating DC computing components into homogenously or heterogeneously resourced pools.…”

“…A homogenously resourced pool is a server-like node which comprises of computing components of the same type (i.e., CPU, memory, or storage). Such homogenous nodes are subsequently allocated to racks in the composable DC at different scales i.e., rack-scale, pod-scale, or DC-scale [17], [20], [21]. At rack-scale, the DC comprises of many racks; each rack comprises of many homogenously resourced nodes of different resource types; and the resources within each node can only be used in conjunction with resources of other corack nodes to form a logical server.…”

“…Rather, the utilization is enforced virtually on-demand using knowledge of the infrastructure state and of applications' specific demands. Logical disaggregation supports repurposing of the server-centric architecture of traditional DCs to enable a rack-scale composable DC that can support all type of applications [17].…”

“…Additionally, electrical topologies adopt a centralized switch architecture which is somewhat simpler and is well known due to its wide adoption in modern DCNs. However, electrical switches are known to have high power consumption [17]. Furthermore, compared to the high bandwidth communication required between disaggregated computing components in composable DCs, the capacity supported by traditional electrical switches are relatively lower.…”

“…Although, a full mesh physical topology is desirable to interconnect disaggregated components within the rack of a composable DC, adopting a generic design that requires dedicated transceivers on each point-to-point link of the mesh fabric is an overkill as the cost may out-weigh the benefits. Furthermore, inappropriate use of electrical switches in composable DCs can lead to significant increase in total DC power consumption [17]. In this paper, we leverage optical components, optical networking techniques, and technologies to minimize the number of transceivers per node while maintaining full mesh connectivity within each rack in the composable DC.…”