The client functions are divided into the following submodules:
- Data. Functions for serialization and deserialization of data from the JSON format to Exonum binary format
- Cryptography. Functions for calculating hashes, creating and validating digital signatures
- Proofs. Functions for checking cryptographic proofs that are returned by the blockchain, such as the functions for checking the proofs for Merkle and Merkle Patricia indexes
- Blockchain integrity checks. Function for checking the validity of a block (its compliance with consensus algorithm)
A hash of the entire blockchain state is a part of each block (see data storage). This hash is formed using Merkle and Merkle Patricia indexes. Thus, using functions for proof verification, one can verify the commitment of any blockchain data in a block. The block itself can be verified using blockchain integrity verification.
There are two typical use cases for the light client:
- Forming and sending transactions to the Exonum blockchain network
- Forming requests to full nodes of the network (normally, HTTP GET requests) and validation of the responses
In this and next section all functions indicated in italics are the functions implemented in Exonum light client.
- Frontend triggered by an event (for example, a button click handler) decides to create a new transaction
- The transaction data is stored in JSON format. Then the data is converted to the Exonum binary format and digitally signed
- The generated transaction (JSON data + digital signature) is sent to the a full node via an HTTP POST request
Serialization during signing is a necessary step, since all data (including transactions) is stored in Exonum in a custom binary format. This is done for several reasons:
- The binary format is unambiguous, while the same data can have multiple JSON representations (which would lead to different hashes of logically the same data)
- The data stored in the binary format consumes less disk space
- Access to a field in a binary format can be implemented using fast pointer arithmetic; the same operation for JSON data would require multiple reads
- The client forms an HTTP GET request and sends it to a full node in the Exonum blockchain network
- The node forms a response to the request and the corresponding
cryptographic proof and sends both back to the client. The cryptographic proof
includes a block header together with
Precommitmessages that certify its validity, and one or more Merkle paths that links the response to the block header.
- The client, on receiving the response from the blockchain, verifies the structure and validates cryptographic proofs for the response.
- The verification procedure includes checking whether a returned response
is stale. This is accomplished by calculating the median of timestamps recorded
Precommits and comparing it against the local time of the client. If the median time in
Precommits is too far in the past, the response is considered stale, and its verification fails.
- The result of checks is shown in the user interface
In the case user authentication is needed (for example, for data access management), requests can be digitally signed.
Blockchain, as a technology that expands the capabilities of distributed databases, implies the possibility of auditing or validating stored information. This audit can be done in several ways.
In a system with full nodes (nodes that store the full copy of a blockchain), audit can be performed automatically during operation of such nodes. When a new block is received, a full node verifies its compliance with the consensus algorithm and the correctness of transaction execution. Such an "audit by dedicated auditor nodes" can be effective in case it is performed by external independent parties. However, it has a number of drawbacks:
- To perform audit, one needs a full copy of a blockchain and enough computational resources to stay in sync with the blockchain validators
- This type of audit needs to be permanent to ensure the correctness of system state
- In order to start auditing, a full auditor node must reach the current height of the blockchain. This could take long time, especially if the blockchain has substantial transaction throughput
- An end user of the system is forced to trust pair "validators" + "auditors", since he has no way to verify the correctness of blockchain data. As a matter of fact, such blockchain is equal to distributed database from the user's point of view.
One of the ways to address the drawbacks of "audit by dedicated auditor nodes" is to introduce light clients, also known as lightweight clients, thin clients or just clients. For the Bitcoin blockchain, these clients are also known as SPV (simple payment verification) clients. Light clients are programs able to replicate and verify a small portion of information stored in the blockchain. Usually clients verify information relevant to a specific user (for example, the history of his transactions). This verification is possible due to the use of specific data containers in a blockchain: Merkle and Merkle Patricia indexes.
Advantages of this approach are:
- Decreased trust in third parties: verification of all data returned by a blockchain in response to queries is performed by the user himself on his machine
- Constant (albeit partial) audit of the system is possible without requiring computational resources comparable to validators in terms of performance. Only the relevant data is audited
- To start or resume the audit of the system, no synchronization period is required
The user still needs to trust the light client developers if the client has closed source. Alternatively, the user should perform an audit of light client code to remove the necessity of trust to third parties completely.
The presence of light clients does not mean the absence of auditor nodes, since their tasks are different. Light clients verify particular user's data, while auditor nodes verify a blockchain as a whole.
Light client security could be compared to TLS security checks embedded into web browsers. It is not a direct substitute to auditing performed by auditor nodes, but it provides a measurable degree of security against MitM attacks and maliciously acting nodes that the client may communicate with. At the same time, if light clients cover all blockchain transactions, their collective security can become comparable to the security provided by the auditor nodes.
The presence of light clients in a blockchain-based system leads to certain difficulties during development:
- Backend developers should agree with client developers on API requests and the format of cryptographic proofs (in fact, blockchain data model)
- Any changes in blockchain data model should be accompanied with relevant changes in the logic of proof verification performed by light clients
- Since the light client substantially expands an access to Exonum REST endpoints with cryptography, it may be necessary to create multiple light clients and continuously support their codebase
The first two problems above can be overcome with the aid of data schema, stated in a language independent format (see the Exonum roadmap).
Despite the complexity of the development, the presence of light clients in a blockchain-based system is the only practical way to largely remove the necessity of trust to third parties.