Читать книгу From Traditional Fault Tolerance to Blockchain - Wenbing Zhao - Страница 62

The normal operation of the protocol is shown in Figure 2.17.

Figure 2.17 Normal operation of the sender-based logging protocol.

The protocol operates in three steps for each message:

1 A REGULAR message, <REGULAR,seq,rsn,m>, is sent from one process, e.g., Pi, to another process, e.g., Pj.

2 Process Pj determines the receiving/execution order, rsn, of the regular message and informs the determinant information to Pi in an ORDER message <ORDER, [m], rsn>.

3 Process Pj waits until it has received the corresponding acknowledgment message, <ACK, [m]>, before it sends out any REGULAR message.

The original sender-based message logging protocol [13] was designed for use with unreliable channels. Since we have assumed the use of reliable channels, one might wonder if the third step in the protocol is still necessary. The answer is yes because transport-level reliability does not necessarily lead to application-level reliability, as we have argued in section 2.3.1.2. If a process sends the ordering message to a process and another regular message to a different process, and node on which the process runs subsequently crashes, the ordering message might not be delivered to its intended target successfully while the regular message might.

Furthermore, in the original sender-based message logging protocol [13] , the regular message and the ordering message must be retransmitted after a timeout before the expected acknowledgment message is received. With the use of reliable channels, such proactive retransmission becomes unnecessary because the only scenario in which a retransmission is necessary is when a process fails, in which case, the retransmission will be triggered by the recovery mechanism (more in section 2.3.2.3).

The use of a mature reliable communication protocol such as TCP in distributed applications is more desirable because the application developers can focus on the application logic and application-level messaging reliability without worrying about issues such as achieving high throughput and doing congestion control.