CAJ | 학술논문

在分布式存储系统的数据中心，数据块通常都是以条状的方式分散存储在不同的服务器中。当客户请求数据时，多个服务器将会同时响应，将分散存储在服务器中的数据传输给客户。当传输的数据量超过交换机缓冲区大小就会在数据中心网络中发生TCP Incast现象。为改善数据中心网络发生TCP Incast吞吐量崩溃问题，提出基于紧迫度分组调度算法。通过在传输过程中将并发传输的数据流分组，并且按照紧迫度顺序对其传输，将并发传输数据进行分组调度传输，实现并发数据的串行传输。仿真实验表明，该算法大大降低了TCP Incast发生率，在服务器数量增加的情况下依然可以保证吞吐量没有出现大幅下降现象。基于紧迫度分组调度算法在大多数用例中可以避免Incast问题，并且提升网络传输性能。
재분포식존저계통적수거중심，수거괴통상도시이조상적방식분산존저재불동적복무기중。당객호청구수거시，다개복무기장회동시향응，장분산존저재복무기중적수거전수급객호。당전수적수거량초과교환궤완충구대소취회재수거중심망락중발생TCP Incast현상。위개선수거중심망락발생TCP Incast탄토량붕궤문제，제출기우긴박도분조조도산법。통과재전수과정중장병발전수적수거류분조，병차안조긴박도순서대기전수，장병발전수수거진행분조조도전수，실현병발수거적천행전수。방진실험표명，해산법대대강저료TCP Incast발생솔，재복무기수량증가적정황하의연가이보증탄토량몰유출현대폭하강현상。기우긴박도분조조도산법재대다수용례중가이피면Incast문제，병차제승망락전수성능。
In a data center of distributed storage system,the data blocks are usually stored in stripe as SRUs in these different servers. When multiple data senders simultaneously communicate with a single receiver,namely in many-to-one communication pattern,the burst data overloads the receiver’s switch buffers,which leads to TCP throughput collapse,that’s so-called TCP Incast problem. In order to improve the throughput collapse problem caused by TCP Incast in the data center networks,propose the algorithm Based on Pressing de-gree Group Scheduling ( BPGS) . Transfer the flows by grouping the flows and ordering them by the pressing degree. Simulation results show that the algorithm reduces the incidence of TCP Incast greatly,even when the number of servers increase,this algorithm can also en-sure the throughout. BPGS can avoid the TCP Incast in most cases,what is more,it can also improve the network transmission perform-ance.