ETYANG LENNAH IPALE –SCT321-C014-2789/2017 JKUAT-ELDORET CAMPUS
MIT 3109: CLOUD COMPUTING
A review on the Research Paper: Distributed Snapshots: Determining Global
States of Distributed Systems
What the paper communicating about.
The paper presents an algorithm that is describing a process which shows a global state of a distributed system during its computation. It explains that many Distributed Systems face a problem of not being able to detect global states using the best algorithm. As a way of communication, processes send and receive messages and the process is able to record its own state and messages that are sent and received. In order to determine this state, a process enlists other processes that cooperate and also records their own states before sending them. All these processes are not able to record these states exactly at the same time unless they have access to a common clock.
What they did for better understanding.
The Authors have cited some properties of a stable algorithm and described very crucial number of problems that can be solved by this algorithm and also the process of devising an algorithm so that the distributed system can be able to establish whether it is a stable property. They have described deadlock as well as termination problems and also cited other authors that have proved that many Algorithms which have been published are neither correct nor practical despite having published a number of algorithms.
What they have used.
They used a model of a Distributed system that has a known number of processes and channels. This model is described by a graph that is labeled with processes and channels. Channels are error free and deliver messages in the order they were sent. The delay experienced by a message in a channel is known. To illustrate this, they considered a system which consists two processes p and q as well as channels c and c’. The system has a token which is passed from a process to the next. They also explained the algorithm that works as follows: each is able to record it’s own state and also two processes that record the channel state. The algorithm is put on a computation, hence must concurrently run with it without any alterations.