Optimal VDC Service Provisioning in Optically Interconnected Disaggregated Data Centers

Abstract

Abstract-Virtual Data Center (VDC) is a key service in modern Data Center (DC) infrastructures. However, the rigid architecture of traditional servers inside DCs may lead to blocking situations when deploying VDC instances. To overcome this problem, the disaggregated DC paradigm is introduced. In this paper, we present an Integer Linear Programming (ILP) formulation to optimally allocate VDC requests on top of an optically-interconnected disaggregated DC infrastructure, aiming to quantify the benefits that such architecture can bring when compared to traditional server-centric DCs. Moreover, a lightweight Simulated Annealing (SA)-based heuristic is provided for scenarios where the ILP scalability is challenged. The obtained numerical results reveal the substantial benefits yielded by the resource disaggregation paradigm.

Future Energy Efficient Data Centers With Disaggregated Servers

“…Constraint (14) ensures that the total number of served VMs is within an acceptable predefined percentage of the incoming VMs requests according to the service level agreement value. The total number of served VMs, depends on the outcomes of constraints (15)- (21) collectively; if all these constraints yield a value of 1, then is 1, otherwise is 0, where and are binary variables indicating if the processing, memory and IO requirements of request are, respectively, served or blocked.…”

“…In this approach, servers' resources are separated into discrete pools of resources that are mixed and matched in real time to create differently sized and shaped systems. This technique brings a new server vision for data centers and motivates a plethora of potential new applications and services [14].…”

On the benefits of resource disaggregation for virtual data centre provisioning in optical data centres

“…Such a feature is especially beneficial when provisioning complex cloud services, such as VDC instances where an optimal utilization of the computing resources becomes paramount. Thanks to the better resource assignment that disaggregated DCs allow when allocating VMs, a higher number of VDC instances can be deployed on top of a shared physical infrastructure [19]. Additionally, the total amount of computing resources to serve the same VDC request set can be lowered when compared to legacy server-centric architectures.…”

“…As a previous work, in [19] we analysed the benefits of optically interconnected disaggregated DCs in terms of increased service acceptance when compared to server-centric DC infrastructures. For this purpose, we proposed exact and heuristic mechanisms aiming at maximizing the number of service requests mapped over a shared DC infrastructure.…”

“…Similarly as in the server-centric case, we also introduce a secondary term (second term in the summation) to prioritize solutions leading to the minimum total number of employed wavelength channels in the DCN. As for the constraints, constraints (12)- (19) have the same purpose as their homonyms in the server-centric formulation (Eq. (2)-(9)), accounting for the rack dimension instead for particular servers, while constraint (20) serves to discriminate if a rack is being employed or not.…”