Clustering Data Log CITA Blockchain Performance Monitoring with Mean Shift Method

• Main Authors: Ubaya, Huda, Stiawan, Deris, Suprapto, Bhakti Yudho, Perdana, Muhammad Iqbal
• Format: Conference Proceeding
• Language: English
• Subjects: bandwidth; Blockchain; CITA; Clustering; Clustering algorithms; Data processing; Feature extraction; Mean Shift; Microservice architectures; Multimedia systems; Real-time systems; Scalability; Silhouette Score
• ISSN: 2837-5203
• DOI: 10.1109/ICIMCIS60089.2023.10349026

CITA is a high-performance enterprise blockchain system with six microservices (RPC, Auth, Consensus, Chain, Executor, and Network) that exchange information through Message Bus. There is a dataset in the form of performance logs from CITA in the format of CPU, Memory, Network, PReads, PWrite, and VMem taken in real-time using the Performance Monitoring Framework for Blockchain. This framework has the advantage of lower overhead, more details, and better scalability than other performance monitoring approaches, but there is no clustering feature in the monitoring results. Thus, it is necessary to include a clustering feature so that users know the performance directly without the need to analyze it more deeply. This research uses the Mean Shift algorithm, which does not require the number of clusters for the initial process but only determines the maximum distance of centroid displacement through calculation. The number of clusters estimated is 2 clusters, where CITA produces 7 cluster samples with cluster evaluation results with the highest average silhouette score from CITA, which is 0.555 in the Network-Received Network-Sent feature pair. So, it is known that the Mean Shift clustering algorithm produces goodness of fit results from clustering on the CITA Blockchain performance log, which means that the clusters are well separated and can be clearly distinguished.