Risk-aware Data Replication to Massively Multi-sites against Widespread Disasters
Shinya Matsumoto, Takaki Nakamura, Hiroaki MURAOKA · 2011
Immediately after a widespread disaster, critical information such as census registers should be available in the disaster-affected area to confirm people’s safety. However, when a widespread disaster actually happened, such information was not available in the Great East Japan Earthquake of 2011 even though the information were protected in safe and remote site over the Internet. This is because the disaster damages both data and internet access of the area. To enable the information available even immediately after a widespread disaster occurs, we propose Risk-aware Data Replication (RDR), which replicates data to safe and nearby sites in the area without the wide area network. It evaluates the site-to-site risk of data loss, and seeks low-risk sites from massively multiple sites using mathematical programming approach. The result of a simple earthquake simulation shows that RDR improves data availability to more than 10% compared to the case of random selection of replication sites and the computation time is less than several seconds even under the 1350 sites. Our contribution is to provide an organized, effective, and feasible way to decide replication sites based on their disaster risks under massively multiple sites environment.