Comparison of Performance Between Kafka and Raft as Ordering Service Nodes Implementation in Hyperledger Fabric

Abstract

One of well-known Hyperledger platform named Hyperledger Fabric, which was first
developed by IBM and Digital Asset, is a blockchain framework with execute-ordervalidate architecture. With this architecture, transactions on Hyperledger Fabric become
deterministic. One of the nodes that plays an important role in the process is ordering
service node. It plays a role in maintaining channel configuration and create a block of
transaction. In its implementation, Hyperledger Fabric consists of three ordering service
node types, namely Solo, Kafka, and Raft. Solo does not have a consensus algorithm and
is only used during the development of blockchain networks while Kafka and Raft have
(crash fault tolerant). This paper simulates Blockchain Network with Kafka and Raft as
implementation of the ordering service node. From the research obtained, it is known that
Raft is more superior than Kafka in terms of success rate and throughput rate when
conducting invoke transactions due to its simpler framework (Raft has the Raft consenter
which was created directly from ordering service nodes) than Kafka (which requires
Kafka brokers and Zookeeper Ensembled). However, in the process of querying the
transaction, Kafka is superior to Raft since the throughput rate is bigger than Raft.