The recommended setting for the server-side MuleFetchMessageSize parameter differs depending on the speed of the network:
- A high speed network connection between the client and server. For example, a 1GE /10GE LAN.
Because of the high network speed, the data is transferred quickly between the server and client without introducing any significant delay.
In this case, the default parameter values are:
- MuleFetchMessageSize 100KB
- FetchCount = 32K
You can use the default MuleFetchMessageSize of 100KB or try increasing it to 1MB for better performance. While increasing the value of the MuleFetchMessageSize parameter reduces the number of data transfer connections between the client and server, it adds overhead by filling the buffer on the server. Therefore, the exchange of several small messages over a high speed network performs better than adding the additional overhead of filling a single large server message buffer while the client waits for the data.
The average performance degradation as compared to the Legacy is:
- 1.5x when using the default MuleFetchMessageSize of 100KB.
- 1.3x when using the recommended MuleFetchMessageSize of 1MB.
The performance improvement when using the recommended MuleFetchMessageSize is 0.2x.
- A low speed network connection between the client and server.
Because of the low network speed, the transfer of data over the network must be reduced. Since the clients will receive the data after a longer period, it is best to fill the server message buffer instead of making numerous data transfer connections between the client and server over the network.
In the case of a single query execution, set the MuleFetchMessageSize parameter to 10MB.
In the case of high concurrency queries, set the MuleFetchMessageSize parameter to 1MB in order to avoid server thrashing.
The average performance degradation as compared to the Legacy is:
- Approximately 20x when using the default MuleFetchMessageSize of 100KB.
- Approximately 2x when using the recommended MuleFetchMessageSize 10MB.
The performance improvement when using the recommended MuleFetchMessageSize is 18x.