DAI 2023-2024 - Java IOs benchmarking - Alexandre Philibert

[AlexandrePhilibert]

Repo

Which type of stream is the most efficient for each use case?

Buffered

Why is it more efficient than the other types of streams?

Because we do not do a syscall for each byte, thus reducing the overhead of each .write() or .read() calls

What is the difference between binary data and text data?

One is human readable

What is a character encoding?

A character encoding specifies how to "translate" bytes into text and the other way.

Why is this methodology important?

Understanding how to best use the tools at our disposal is important

Text Data

Size	Buffered Write	Unbuffered Write	Buffered Read	Unbuffered Read
1B	191990ns	374988ns	323098ns	787075ns
1024B	678495ns	3211851ns	705616ns	2794606ns
1MB	14ms	66ms	21ms	50ms
5MB	103ms	175ms	90ms	152ms

Binary Data

Size	Buffered Write	Unbuffered Write	Buffered Read	Unbuffered Read
1B	411293ns	299945ns	134283ns	156695ns
1024B	105153ns	98899ns	426655ns	1368098ns
1MB	1ms	2ms	20ms	943ms
5MB	5ms	9ms	96ms	4628ms

© ChatGPT for the tables

[ludelafo]

Thank you for the mention of the tables' source! Great usage of ChatGPT.

I would add that this scientific methodology to benchmark our applications can ensure the performance are still good over time for the use cases we are managing.

EDIT: For text files, do not forget to specify the charset of the file (while reading and writing!). It is more explicit and you can avoid many issues in the future.

One of the constructors for the FileReader et FileWriter classes allows to specify the charset to ensure the characters are interpreted correctly by your application:

• https://docs.oracle.com/en/java/javase/17/docs/api/java.base/java/io/FileReader.html
• https://docs.oracle.com/en/java/javase/17/docs/api/java.base/java/io/FileWriter.html