amitkale@clusterfs.com
2007-Jan-25 23:38 UTC
[Lustre-devel] [Bug 10968] Client VFS operations statistics for /proc
Please don''t reply to lustre-devel. Instead, comment in Bugzilla by using the following link: https://bugzilla.lustre.org/show_bug.cgi?id=10968 Requirement #1 producing graphs-> Would following steps achive the objective of producing graphs to analyze some behavior? We''ll code these steps in a script. 1. Run a set of test cases 2. For each test case collect statistics of VFS operations performed by processes spawned by the test. 3. Plot a graph for each of these test cases. Y axis will indicate the number of calls for a VFS operation. X axis will be of test configurations. Each worm in the graph will represent one VFS op. 4. A wrapper script will take input of a path to a test script to be run for each instance. Each later parameter will contain arguments to be passed to the test script for each instance. The wrapper script will run given test script for each parameter and collect statistics for the same. This will require a utility (described in point number 5 below) written in C. 5. A utility written in C - Runs a program with given parameters.It first spawns a thread, adds that task to /proc tracker after clearing it first. Then execs the given program. It waits till the program terminates.