Scaling Hyperledger Fabric Using Pipelined Execution and Sparse Peers
share
Many proofs of concept blockchain applications built using Hyperledger Fabric, a permissioned blockchain platform, have recently been transformed into production. However, the performance provided by Hyperledger Fabric is of significant concern for enterprises due to steady growth in network usage. Hence, in this paper, we study the performance achieved in a Fabric network using vertical scaling (i.e., by adding more vCPUs) and horizontal scaling (i.e., by adding more nodes) techniques. We observe that network scales very poorly with both of these techniques. With vertical scaling, due to serial execution of validation commit phases of transactions, the allocated vCPUs are underutilized. With horizontal scaling, due to redundant work between nodes, allocated resources are wasted though it is utilized. Further, we identify these techniques to be unsuited for dynamically scaling a network quickly to mitigate an overload situation, and hence, it results in a 30 in the performance. To increase the CPU utilization and hence the performance, we re-architect Fabric to enable pipelined execution of validation commit phases by introducing dirty state management using a trie data structure. Additionally, we facilitated the validation phase to validate transactions in parallel by introducing a waiting-transactions dependency graph. To avoid redundant work performed between nodes and to quickly scale up a network, we propose a new type of peer node called sparse peer, which selective commits transactions. Overall, we improved the throughput by 3x and reduced the time taken to scale up a network by 96
READ FULL TEXT