Fujitsu Laboratories Ltd has developed a technology that fast tracks transaction processing for Hyperledger Fabric, one of the Hyperledger blockchain frameworks hosted by The Linux Foundation.

Blockchain is a technology that creates a shared ledger system that, without a centralised manager, is highly transparent and reliable while being difficult to falsify, by requiring the parties involved to mutually verify the accuracy of the transaction data and preserving it in a chain format. It is expected to have applications in a variety of fields, particularly in finance.

The open source blockchain framework Hyperledger Fabric, being developed through Hyperledger, in which Fujitsu Ltd is a premier member, uses a consortium-type structure in which the number of participants is limited, and is being trialed for use in a wide variety of fields, including finance, supply-chain management in manufacturing, data conversion of insurance policies, real estate contracts, license management and energy transactions.

The bottleneck

In blockchain, groups of nodes based on the number of participants form a network and work together through the network to perform a series of processes from executing transactions to validating the legitimacy of transactions. This makes the number of transactions that can be executed per unit of time limited by communication bottlenecks through the network, compared to previous centralised systems. This poses a challenge to applying the technology to things like online transaction systems, which demand high performance, including the ability to immediately process large volumes of transactions.

20170809_Fujitsu_01 Figure 1: Transaction processing on the blockchain. While consensus is formed between participating nodes, transactions are processed by applications reading and writing data on the shared ledger, with safety ensured by linking the transaction data in chain format to be managed.

Using a proprietary analysis technology, Fujitsu Laboratories learned that, under network conditions in which a response time of about 64ms or less is required, such as the case where a consortium-type blockchain is operated at multiple locations within Japan, communications between the applications and the blockchain platform during transaction processing are the primary cause of bottlenecks. Based on this knowledge, Fujitsu Labs developed two technologies to improve transaction performance speed by reducing the number of communications between the applications and the blockchain platform.

Speeding up transactions

Differential update state (DUS): When processing transactions on the blockchain, a commonly used method is to retrieve the specified data, then handle the computational processing in the application before writing it back to the blockchain platform. DUS functionality executes only differential computations on the specified data, in one processing action on the blockchain platform, while Fujitsu additionally reduces the number of computations directly linked with the number of communications.

Compound request (CR): CR functionality aggregates multiple processes to send to the blockchain platform for batch execution. This improves efficiency of both processing and communications. The functionality maintains accuracy by rewinding to the origin point of the batch execution if a partial error occurs in the aggregated processes and reprocessing.

20170809_Fujitsu_02 Figure 2: Example showing the reduced number of communications due to the DUS functionality and the CR functionality.

Fujitsu Labs has implemented this technology in Hyperledger Fabric v0.6.1 and measured transaction performance on a blockchain platform consisting of four servers. Whereas the previous method could handle 500 transactions per second, the company achieved 1,350 transactions per second using the new technology, an improvement of about 2.7 times. This makes Hyperledger Fabric framework applicable to online transaction systems that demand performance in excess of 1,000 transactions per second, such as those demanded by financial institutions.

Fujitsu is planning to commercialise the technology during fiscal 2017.