Differences

This shows you the differences between two versions of the page.

--- how_we_measure [2020/12/19 17:23] – created senioradmin
+++ how_we_measure [2021/01/16 22:35] (current) – [Measuring with smem] senioradmin
@@ Line 5: / Line 5: @@
 There are several values which can be used to determine the "lightweightness" and complexity of a program.
-* Size of the source code
+  * Size of the source code
-* Size of the compiled binary
+  * Size of the compiled binary
-* Lines of Code
+  * Lines of Code
-* the programming language which was used
+  * the programming language which was used
-* RAM usage
+  * Memory usage
 We think, the first three values are giving only limited indications on how lightweight a  program is.
@@ Line 15: / Line 15: @@
 The programming language is a strong indicator - a program which was programmed in a language that translate directly to machine code, made by a compiler will be more efficient and lightweight as a program which was programmed in a language which bust be run by an interpreter or bytecode interpreter.
-An even stronger indicator is memory usage. So we will measure the memory usage oa a program to give clear recommendations. we do this by using the program [[https://www.selenic.com/smem/|smem]].
+An even stronger indicator is memory usage. So we will measure the memory usage of a program to give clear recommendations. we do this by using the program [[https://www.selenic.com/smem/|smem]].
+===== Measuring with smem =====
+When it comes to memory measurement there are different terms with different meanings. Most standard Unix tools are reporting the "resident set size" ([[glossary#RSS|RSS]]) - which roughly is the memory usage of a program plus the shared libraries it needs. Because it includes the whole shared memory it shares with other programs this isn't a fair representation of memory a program really needs.
+The "unique set size" ([[glossary#USS|USS]]) on the other hand underestimates the programs memory footprint, because it reports only the memory needed by the program binary only, without any shared library it might need.
+The "proportional set size" ([[glossary#PSS|PSS]]) smem reports is the memory the program plus the '"fair share" of each shared area to give a realistic measure.' That means, of course, the program's PSS will shrink the more programs with the same shared libraries will run.
+We will measure all programs under the same condition (a machine with 2 GB RAM), by running as few as background processes as possible and only those which are needed for the program - obviously we have to run X11 to measure the memory usage for a graphical program which runs under X11 (using a resolution of 1024x768). As it is default on Linux, the **unit used is kilobyte (kB)**