Skip to content

Latest commit

 

History

History
109 lines (70 loc) · 3.66 KB

INSTALL.md

File metadata and controls

109 lines (70 loc) · 3.66 KB

IOSIG - INSTALLATION GUIDE

This file is written using Markdown. It can be viewed in a Markdown viewer or converted to a single-page HTML document.

Software dependencies

Download IOSIG and setup the environment variables

$ tar zxvf iosig-1.4.tar.gz
$ export IOSIG_HOME=./iosig-1.4
$ mkdir ~/.iosig
$ mkdir ~/.iosig/data
$ export IOSIG_DATA=${HOME}/.iosig/data

NOTE: we suggest you to add these two environmental variables to your .bash_profile or .bash_rc files. The variable IOSIG_HOME is used to locate the IOSIG tools, including the library and the analyzer. The variable IOSIG_DATA is used to store the trace file.

Compiling the trace library.

$ cd ${IOSIG_HOME}/src/collect
$ vim Makefile

Edit the first two lines of Makefile. Set MPIPATH to the source code of MPICH2, which has been configured ($ ./configure). Set INSTALL_DIR to where you want to install the IOSIG library.

$ make
$ make install

Using examples

$ cd ${IOSIG_HOME}/src/examples

There are several separated directories, each of which is an example of using IOSIG.

  • execution_trace
  • IOR
  • posix_io_c
  • mpiio_c
  • mpiio_fortran

The directory names tell what the examples are. In each example, there are two Makefiles:

  • Makefile
  • Makefile.iosig

The first is the default Makefile to compile the example program. The second also enables IOSIG during building the program. Using Makefile.iosig to build and run the program will generate the IO traces.

By comparing Makefile and Makefile.iosig, you get the idea of how to link IOSIG library with your program.

Using IOSIG tracing library

To use IOSIG, just add the following compilers flags to your gcc or mpicc command line. If you have difficulty on how to build your program, please check the examples following the instructions of last section.

IOSIG_CFLAGS   = -w -finstrument-functions
IOSIG_LDFLAGS  = -Wl,-wrap,fopen,-wrap,fopen64,-wrap,fclose,-wrap,fread,-wrap,fwrite,-wrap,fseek,-wrap,open,-wrap,close,-wrap,read,-wrap,write,-wrap,lseek,-wrap,lseek64,-wrap,open64 -L${IOSIG_HOME}/src/collect -liosig

IOSIG_CFLAGS is to be added in the compiling stage. IOSIG_LDFLAGS is to be added in the link stage. Please check the examples for how to do this.

For more examples on how to use IOSIG to profile parallel I/O benchmarks and applications, check the GitHub project: https://github.com/yinyanlong/profiling_parallel_io.

Using IOSIG analyzer

In this current version, you need to run the analysis manually for each trace directory in $IOSIG_DATA. For example:

$ cd ${IOSIG_DATA}
$ cd username_1386547880_userApp
$ source ${IOSIG_HOME}/src/analysis/plot.sh

The results will be generated in a directory called result_output in the directory username_1386547880_userApp. Do not modify or delete the content of the directory since the web dashboard needs to read and display the results. Next section shows how to use the web dashboard.

Using the web dashboard

You need to download and install the Flask python package, which is available at: http://flask.pocoo.org/.

After that, simply follow the following steps:

$ cd ${IOSIG_HOME}/src/webapps
$ python webapp.py

Then, open your browser and access http://hostname:5000/. Here, hostname is the machine where you run python webapp.py.

NOTE: in this version, the web dashboard can only display the trace that has been analyzed. Future version may make all analysis procedures automatic.