My paper of choice was "A solution based on the MPP'S to storage mass data" by zhiyu Chen liufu Song. The discussed the problem of storing increasing amount of data, and the possible solution for the shortage of physical disc space availability. Two major approaches were identified, one is data compression and the other one is Massively Parallel Processing technology (MPP).
Since compression offers fair amount of success in space usage optimization solutions, as the amount of data is exploding, this approach can not maintain the competitive edge against distributed data.
In their MapReduce (introduced by Google, currently Hadoop MapReduce by Apache is really popular) solution, they explained, how can a network system can function as a cheap and efficient data storage solution. They basically using nodes instead of distributed computing to do distributed data storing. They hypothesized a master-mirror setup for loss prevention, where the mirror systems shall hold the backup information, in case of failure on the master side.
This article have a good idea, however the paper itself is poorly structured and in many places the wording is not correct. Also it lacks in depth insight into many fundamental problems with the proposed approach. They fail to address the problem, where multiple failures occur at the same time, undesired segment disclosures happen, or when data corruption happens. They also hypothesized a static network settings, where no changes in network topology is made and further assume, that every node is trustworthy. Also, there is no mentioning of human interaction or interference, and also lacks the coverage of data synchronization and load balancing issues. The paper does not cover large scale problems, where MapReduce could introduce better solutions.
Thanks for reading
--
Balazs Makany
No comments:
Post a Comment