Research On Key Technologies For Big Data Movement In High-performance Networks

Today’s large-scale scientific application are predominantly collaborative in nature,pulling together national or even globally distributed resources and coordinating among multiple participating sites to accomplish missions that are simply impossible individually,and data-centers of scientific research,business and high-performance computing need the fast and reliable network.Traditional Internet is not adequate to meet these stringent network re-quirements.In recent years,high-performance networks(HPNs)based on circuit-switching or MPLS/GMPLS tunneling techniques or SDN have emerged as a promising solution to solve this problem.HPNs are computer networks with high bandwidth and bandwidth pro-visioning and their significance has been well recognized in the broad science and network research communities.It is urgent task for researchers on how to improve the utilization of network resources and meet diverse user requests.There are two aspects of data transfer,i.e.bandwidth scheduling and transport control,which have been the focus of research since HPNs came into being and there are remark-able accomplishments in each area.However,the majority of existing work in bandwidth scheduling is focused on instant scheduling.Periodic scheduling for multiple reservation requests has attracted more attention in recent years due to its advantages in improving net-work utilization and meeting diverse user requests from a global perspective.The transport control also plays an important role for the effective utilization of bandwidths provided by high-speed links because even after dedicated channels are successfully provisioned,the ac-tual data transfer performance still largely depends on the specific transport protocol being used.Besides,bandwidth scheduling and transport control have been traditionally treated as two separated topics,but in fact they are not completely independent of each other,but have intricate interactions.This dissertation focuses on periodic bandwidth scheduling and explores an integration of bandwidth scheduling and transport control for periodic scheduling.First we proposed new solutions for two periodic scheduling problems and then formulated two new problems of periodic scheduling and proposed an algorithm for each of them,and last we explored an pe-riodic scheduling problem with an integration with transport control.The main contribution of this dissertation is given as follow:(i)Periodic Scheduling of Fixed-slot Bandwidth Reservations,which is defined as M-FSFB and is shown to be NP-complete,and its goal is to maximize the number of satisfied fixed-slot bandwidth reservation requests.The user request specifies five parameters:a source node and a destination node,a required bandwidth and a time slot defined by a pair of s-tart and end time.The distribution of time-slots of the reservation requests falls in one of three categories:1)identical time-slots,2)partially overlapped time-slots,and 3)complete-ly disjoint time-slots.In this dissertation we first design a minimum resource occupation algorithm for a special type of M-FSBS with identical slots,referred to as MinRO-IS,and further propose a generalized version of MinRO with arbitrary slots.We also design four greedy algorithms,namely minimum bandwidth occupation(MinBO),minimum bandwidth and slot occupation(MinBSO),minimum slot and hop occupation(MinSHO),and minimum bandwidth and hop occupation(MinBHO),for performance comparison.Extensive simu-lation results illustrate that both MinRO-IS and MinRO have superior performance over the existing algorithms in the literature and the other four greedy algorithms in comparison.(ii)Periodic Scheduling of Bandwidth Reservations with Deadline Constraint,which is de-fined as M-DCBS and we shown the NP-completeness of this problem.The user request has four parameters:data size,the maximum bandwidth of Local Area Network,the earliest possible data transfer start time and the deadline time.In the literature there is a polynomial time RRA algorithm to accommodate as many reservation requests as possible between a pair of end sites while minimizing the total time needed to complete the data transfers.The drawback of RRA is that there is too iterations and the second objective may yield to the first one.We propose a Maximum User Number Resource Reservation Algorithm(MUNRRA)and extensive simulation results show that MUNRRA exhibits a superior performance over RRA in terms of scheduling radio and execution time and has a small average transfer time.(iii)Periodic Scheduling of fixed-bandwidth floating-slot under deadline constraint,which is a new problem.We formulate a periodic bandwidth scheduling problem to maximize the number of satisfied user requests for fixed-bandwidth floating-slot under deadline constraint on a network path,referred to as FBFS-DC.This service model specifies four parameters:a fixed bandwidth,a time duration,an earliest possible start time and a deadline.We prove that this problem is NP-complete and propose a bandwidth scheduling algorithm based on product of bandwidth and slot,referred to as Product-BS and also design three heuristic algorithms for performance comparison,which employ slot-oriented scheduling(Slot-BS),bandwidth-oriented scheduling(BW-BS),and First Come First Serve scheduling(FCFS-BS),respectively.Extensive simulation results illustrate that Product-BS has a superior performance over the other three algorithms in terms of scheduling success ratio.(iv)Periodic scheduling of deadline-constrained variable slot-bandwidth reservations,which is also a new problem proposed in this dissertation.We formulate this periodic bandwidth scheduling problem to maximize the number of satisfied user requests for variable slot-bandwidth reservation under deadline constraint on a network path,referred to as VSBR-DC.Each of the user request is represented by dada size,the maximum bandwidth of local area network,the earliest possible start time and the latest transfer end time.We show that this problem is NP-complete,and propose a bandwidth scheduling algorithm based on op-timal scheduling order and allocation strategy,referred to as OSOAS-BS.and also design three algorithms for performance comparison,which employ deadline-oriented schedul-ing(Deadline-BS),availability-oriented scheduling(Availability-BS),and First Come First Serve(FCFS-BS),respectively.Extensive simulation results show that OSOAS-BS has a superior performance over the other three algorithms in terms of scheduling success ratio.(v)Periodic Scheduling with the throughput estimation of end-host,which is also a new problem proposed in this dissertation.We formulate a periodic bandwidth scheduling prob-lem to maximize the number of satisfied user requests for profiling-based floating-window bandwidth reservations under deadline constraint on a network path,referred to as PFWR-DC.We consider multiple user requests in the form of a three-tuple:a floating window size,an earliest possible start time and a deadline,respectively.We prove that this problem is NP-complete,and propose an integrated bandwidth scheduling algorithm based on the prod-uct of floating-window size and dynamic transport throughput,referred to as ProductWT-BS.We also design five heuristic algorithms for performance comparison,which employ duration-oriented scheduling(Dur-BS),throughput-oriented scheduling(TP-BS),deadline-oriented scheduling(DL-BS),availability-oriented scheduling(Ava-BS),and First Come First Service scheduling(FCFS-BS),respectively.Extensive simulation results illustrate that Product WT-BS has a superior performance in terms of scheduling success ratio over the other five algorithms designed for performance comparison.