分散式異質性伺服器之多重資源動態負載平衡

Abstract

由於網際網路的蓬勃發展，使用網際網路服務的人口迅速地增加，因此必須同時使用更多的服務伺服器同時提供服務，來滿足大量的服務需求。同時使用多台伺服器若不能有效地分配負載，不僅造成系統利用度的低落，還會導致服務品質的不佳。為此已有很多研究被提出來解決這方面的問題。依據不同的需求，負載平衡的架構可大致被分為調度器為主、DNS為主、使用者為主，以及伺服器為主的四大架構，且皆具有各自的優缺點。過去在地理分散的服務伺服器負載平衡方法上，服務伺服器為了降低狀態變化的頻率，通常會設定一負載緩衝區間，來降低因服務伺服器狀態變化而產生的網路封包，然而卻會造成各服務伺服器負載震盪的現象。此外傳統方法通常假設各伺服器皆具有相同的負載能力，且通常只考慮服務請求的一種資源消耗，然而事實上，各伺服器間往往不見得都具有相同的負載能力，而各服務請求也不只需要消耗一種伺服器的資源，因此只考慮一種資源消耗，將造成系統瓶頸發生在少數的資源上，導致整體的系統使用率不佳。為此，在地理分散的負載平衡架構上，我們採用Random Early Detection (RED)的機制，來更有效地決定伺服器的狀態，降低服務伺服器負載震盪的幅度。此外在地理集中的負載平衡架構上，我們提出一分散式的市場機制負載平衡方法，不僅考量各伺服器間的異質性，也考量各工作的不同資源需求。透過模擬，我們證明透過RED的概念，對於地理分散的伺服器而言，我們可以有效地降低各伺服器負載震盪的幅度，而對於叢集內的伺服器而言，分散式的市場機制負載平衡方法，不僅可以有效地提高服務伺服器間，以及服務伺服器內部資源的負載平衡性，且即使伺服器的異質性提高，以我們的分散式市場機制負載平衡方法仍可以維持良好的負載平衡性，達到高使用率以及低回應時間等優點。

Due to the progress of the Internet, there are more and more people using Internet services nowadays. In order to satisfy the huge service requirements, using multiple servers to provide different services at the same time is necessary. However, if we can’t effectively divide loads among servers, server utilization could decline and service quality could become uneven. Because of this reason, there are many researches have be presented to solve this problem. The load balancing methods can be roughly classified to four architectures: dispatcher-based, DNS-based, client-based and server-based, and different architectures have their own advantages and disadvantages. The conventional methods of load balancing always set a load buffer range to decrease the state change frequency of a service server in the geographic distributed load balancing architecture, and mostly assume that servers are homogeneous and just consider single resource consumption, such as CPU load. However, the load buffer range would result in load oscillation among servers. On the other hand, servers may not always have the same capacity, and jobs need many kinds of resource requirements. Only considering single resource consumption would cause the system bottleneck to derive from the lack of a small number of resources, and lead to low system utilization. For this reason, in the geographic distributed load balancing architecture, we use the concept of Random Early Detection (RED) to determine the server state probabilistically, and in the cluster load balancing architecture, we present a distributed market mechanism (MM) load balancing method which would consider the server heterogeneity and multiple-resource consumption simultaneously. In our simulation, we show that the oscillation of service server load can be reduced by using the concept of RED in the geographic distributed load balancing architecture. And distributed market mechanism load balancing method can improve the inter-server and intra-server load balancing at the same time, and keep the performance even if the server heterogeneity increasing, achieve high system utilization and low request response time.