Energy-aware Routing Algorithm And Deterministic Bandwidth Guarantee In Data Center

Data center is the center of data transmission,computation and storage.It centralizes various hardware/software resources and business systems,such as Web services,search engines,online contact,online shopping,Map Reduce,etc.In recent years,data center networks have expanded dramatically.With the development of cloud computing and virtualization technologies,data centers have undergone profound changes in terms of composition,structure,functions and application patterns.These characteristics have posed great challenges to the energy supply and cooling of data centers.They have also put forward new demands for the availability and reliability of data centers.In this paper,we do some research on these two aspects.On the one hand,we make use of the link redundancy of network topology to route traffic in subnet while not exceeding the limited threshold of link load,and then sleep/close idle devices,so as to save energy under the premise of ensuring performance.On the other hand,we refine switches,independent nodes on the network topology,to provide deterministic bandwidth guarantee for high-priority data flows.Specifically,we realize multi-queue output structure at flow-granularity and the coordinated scheduling algorithm to ensure the bandwidth allocation.Routing path planning of network traffic is a NP-hard problem.Specifically,the problem we are committed to solving is how to route network traffic with the minimum number of routing devices while satisfying its performance requirement.The energy-aware routing algorithms proposed before mainly focus on two aspects.One is modeling and optimization,and the other is heuristic algorithm design.In this paper,we use CPLEX,an optimization studio,to find the optimal solution.But the running time proves that the time complexity of the optimization is too high to meet the near real-time requirement of data centers.Only simple and efficient heuristics are the practical solutions.However,the development of heuristic energy-aware routing algorithms remains to be explored.Although the few published heuristic algorithms are simple,their judgment operations are redundant and the solutions are not good enough due to the lack of detailed analysis of the problem.So we propose a reasonable energy-aware algorithm framework and design three simple and efficient heuristic algorithms after fully studying the application background.The simulation results show that our heuristic algorithms are effective in energy saving.They can save more than 50% of the network energy under low network traffic load,and can still save 30% when the traffic load is up to 0.5.Further,in view of the research before is only for static traffic context,without considering the dynamic characteristic of real-time traffic,we propose an energy-aware algorithm framework for dynamic traffic context inventively.The heuristic algorithms proposed above for static traffic context can be embedded into the dynamic framework seamlessly to optimize the path planning results.Moreover,traffic shaping and path change are carried out in the dynamic framework to centralize traffic in subnet,and then the redundant ones can be closed.The dynamic framework can ensure the energy saving effect in dynamic scenarios where traffic load is changing.It can also ensure the stability of the selected devices subset,avoid frequent switching of devices and ensure the security and the stability of the devices and the data center network.In the path planning problem of dynamic traffic scenario,we also take the rescheduling of persistent flows into consideration.We want the persistent traffic to be routed on its previous path as much as possible,so as to avoid routing table overwriting,out-of-order situation and rearrangement operation.The simulation results show that with the change of dynamic traffic,the redundant open devices can be shut down for better real-time energy saving effect when the demand for routing devices decreases.When the demand for routing devices keeps the same or increases,the set of selected open device can remain relatively stable.And different heuristics we proposed can support path retention of persistent flows to varying degrees.On the other hand,the diversity of business and the division of Service level put forward different and deterministic demand for the Quality of Service(Qo S)of data center network.We pay attention to the switches,simplify the problem and arrive at a solution.The off-the-shelf physical switches can only have limited queues per output port.Flows converge in the same queue will compete for the shared output bandwidth,and then interfere with each other,unable to achieve deterministic performance.For this problem,we refine the structure and operation of open vswitch(Ov S)from source code to resolve conflict and to ensure the bandwidth guarantee for high priority data flows.Specifically,the multi-queue buffering structures,named PQ(Priority Queuing)at priority-granularity and FQ(Flow Queuing)at flow-granularity,as well as the coordinated scheduling algorithm are designed and realized in Ov S.The application of virtualization technology avoids the superior difficulty and high cost for the update of physical switching equipment.The recovery and reuse of storage make it possible for multi-queue structure at flow-granularity.Finally we flexibly realize the structure and strategy we want by programming.The test results show that the refined Ov S can provide deterministic bandwidth guarantee for priority data flows to realize resource allocation and differentiated services.