分散式系統是使用一些便宜又高效能的電腦緊密的透過網路連結在一起，而彼此間透過網路傳遞訊息來通信。在分散式系統中，訊息廣播是一種常見的群組通信方式。在通信的過程有可能會有訊息流失，或某些電腦的當機，這些都會造成整個系統的錯誤跟無法估計的損失。因此在分散式系統的研究中，許多學者都專注在系統的容錯問題，例如盡早的發現訊息流失
和電腦故障以降低系統災害損失。全域排序(Total ordering)是在分散式系統容錯非常重要的一個課題。「全域排序」是分散式系統的訊息廣播機制，其作用是確保每個處理器在接收其他處理器廣播的訊息順序上一致。如果一個處理器廣播出一個訊息後，則所有正常的處理器都有收到；或是沒有任何一個正常的處理器收到。從訊息上的全域排序可以看出是否有任何訊息的流失或更甚的偵測出崩潰的處理器。本論文研究在Web環境下的一個可靠且高效率訊息全域排序策略，簡稱SPETOS (Starvation ProtEcting Total Ordering Strategy)。SPETOS系統不但可以提供高效率的全域廣播，而且可以避免其他系統常犯的饑餓(Starvation)問題。本論文並用模擬方式去驗証，比較各個全域廣播策略，並得到証實在高頻率廣播狀況下，SPETOS系統表現出較佳的效能且能防止饑餓(Starvation)問題發生。

Distributed systems use inexpensive high-performance computers and can be configured to operate closely by network communication. Broadcast is a very popular kind of group communications in a distributed system.Message loss under communication or some computers failure will result in system fault and unpredictable loss. Therefore in the distributed system, many researchers are focusing on the fault-tolerance issues：such as detecting message loss ,and computer failure as soon as possible to reduce the potential disaster. The total ordering problem is a very important issue for fault-tolerance in distributed systems. "Total ordering" is a broadcast mechanism of the distributed system to insure that all the correct processor deliver the same sequence of message order. If a processor broadcasts a message, then either all the correct processors deliver this message or none of them deliver it. The implementation of total ordering in a distributed system can detect if any message lost ,even more
any processor crash.In this thesis, we propose a reliable ,totally ordered , and high-performance broadcast strategy on World Wide Web environment, called SPETOS.SPETOS system not only can provide high performance total ordering broadcast, but also can avoid starvation problem which often happens in other system. We also use simulation to compare and verify various total ordering strategies ,and we found that our SPETOS system can provide better performance and avoid starvation for a
distributed system when the system is on the status of high broadcast requirement.