Enhancing Leakage Prevention for MapReduce

Next-Generation Intrusion Detection and Prevention System Performance in Distributed Big Data Network Security Architectures

“…However, like most application or service modules within big data frameworks, MapReduce highlights the vast attack vectors that exist in distributed big data systems. MapReduce examples in literature include side channel attacks [19], job composition attacks [20], and malicious worker compromises in the form of distributed denial-of-service (DDoS) or replay attacks [21], Eaves dropping and data tampering [22]. Encryption is a primary countermeasure to secure transmissions and prevent data leaks between big data servers [19].…”

“…Often in Hadoop, data is stored and retrieved from the Hadoop Distributed File System (HDFS). In [19] side-channel attacks are addressed that can occur between MapReduce workers that utilize HDFS for data storage. These types of cybersecurity attacks can target worker nodes to extract valuable information pertaining to MapReduce jobs such as the amount of packet bandwidth.…”

“…This further contributes to successful pattern attacks. The authors proposed a solution to this vulnerability labeled Strong Shuffle that enforces strong data hiding between workers [19]. In contrast to alternative countermeasures such as correlation hiding in [20], Strong Shuffle avoids leaking the number of records accepted by each reducer during MapReduce runtime.…”

“…In contrast to alternative countermeasures such as correlation hiding in [20], Strong Shuffle avoids leaking the number of records accepted by each reducer during MapReduce runtime. Secure plaintext communications is a function of semantically secure encryption in the Strong Shuffle solution [19].…”

“…In [19] data communicated between Hadoop DataNodes and stored in HDFS is encrypted with semantically secure AES-128-GCM encryption. Although the latter helps prevent clear text leakage between MapReduce jobs in Hadoop, encryption in big data environments has limitations.…”