The Short List


OpenMP

g95

Cost

Study Further?

Link

Misc

Vampir/ITAC

n

*9

$200/license

n

http://www.vampir-ng.de/

*1

XMPI
n
y
Free
*17
http://www.lam-mpi.org/software/xmpi/

MPIP

n

y

Free

n

http://mpip.sourceforge.net/

*1

KOJAK

y

y

Free

n

http://icl.cs.utk.edu/kojak/index.html

*6

Paraver

y

*2

Free

y

http://www.cepba.upc.es/paraver/index.html


gprof

y

y

Free

*17

http://www.cs.utah.edu/dept/old/texinfo/as/gprof_toc.html


TAU

y

n, *7

Free

y

http://www.cs.uoregon.edu/research/tau/home.php


PGI CDK

y

*4,*14

*10

y

http://www.pgroup.com/products/cdkindex.htm


OPT

y

*5

*8
n

http://www.allinea.com/index.php?page=74


MPE/Jumpshot
n
*3
Free
n

http://www-unix.mcs.anl.gov/perfvis/software/viewers/index.htm


Paradyn
n
*12
Free
n
http://www.paradyn.org/
SvPablo
y
*2
Free
y
http://www.renci.org/software/svpablo/
VTune
y
*13
*11
n
http://www.intel.com/cd/software/products/asmo-na/eng/vtune/vlin/239145.htm
HPC Toolkit



y
https://domino.research.ibm.com/comm/research_projects.nsf/pages/actc.index.html
Etnus TotalView
y

*15
n
http://www.totalviewtech.com/index.htm
Open|SpeedShop


Free

http://oss.sgi.com/openspeedshop/
Oprofile
n

Free
y
http://oprofile.sourceforge.net/news/
PapiEx
n

Free
n
http://icl.cs.utk.edu/~mucci/papiex/papiex.html
IPM


Free
n
http://ipm-hpc.sourceforge.net/
DEEP/MPI


*16
n
http://www.crescentbaysoftware.com/deep_mpi_top.html
PerfSuite


Free

http://perfsuite.ncsa.uiuc.edu/


Notes:


*1 - Just an MPI profiler

*2 - Seems so

*3 - Seems very likely, but not definite

*4 - Provides an F9X compiler

*5 - Compatible with the "GNU Compiler Suite"

*6 - Technically, KOJAK is not a profiling tool per se. It produces enhanced profiles out of traces. This especially means it does not work very well with long running programs.

*7 - An insider from TAU project emailed us saying that g95 support was going to be added in about 2 months.

*8 - Commercial, price currently unknown

*9 - Compatible with "Intel compilers and GNU compilers"

*10 - http://www.pgroup.com/pricing/ae32L.htm

*11 - $280/academic license, $699/regular license

*12 - Doesn't require code modification or recompilation

*13 - "Compiler and language independent"[2]

*14 - "Fully interoperable with gcc, g77, and gdb" - Website

*15 - License options at http://www.totalviewtech.com/Purchase/Comparisons.php - contact for pricing.

*16 - http://www.crescentbaysoftware.com/deep_mpi_price.html

*17 - Tool already in use - No need for further study





Paraver

http://www.cepba.upc.es/paraver/index.html

Pros:

  • Performance visualization tool
    • performance views: timelines, textual, statistics
    • 3D/2D analysis module
  • Strong documentation + David :)
  • Free (need to get a licence) 
  • Done especially for OpenMP+MPI apps

Cons:

  • Seems complicated
  • We don't have that much time (REU students I mean)
  • Apparently doesn't run on x86


Vampir

http://www.vampir-ng.de/

Pros:

  • Good, stable GUI
  • Provides a large number of view options
  • Shows call trees
  • Powerful
  • Can do binary instrumentation

Cons:

  • Offers no user guidance on analyzing data
  • Only does MPI
  • Solid learning curve
  • $200 per license


TAU

http://www.cs.uoregon.edu/research/tau/home.php

Pros:

  • TAUs profile visualization tool, paraprof, provides graphical displays of all the performance analysis results, in aggregate and single node/context/thread forms. [Website]
  • TAU can generate event traces that can be displayed with the Vampir, Paraver or JumpShot trace visualization tools. [Website]

  • Free license, just have to request a copy trough email
  • OpenMP
  • "Will have support of g95 in two months" (by insider)
  • Already used for WRF
  • lots of documentation

Cons:

  • No g95 Support!
  • Instrumenting is tedious and error prone
  • Steep learning curve
  • Many features depend on other software


gprof

http://www.cs.utah.edu/dept/old/texinfo/as/gprof_toc.html

Pros:

  • Free user license
  • OpenMP
  • Good Tutorial on the web by GNU
  • Quick and Easy to use
  • Already works on our system!

Cons:

  • Doesn't scale well to large jobs
  • No useful MPI or I/O information
  • Don't provide wall-clock time information


KOJAK

http://icl.cs.utk.edu/kojak/index.html

Pros:

  • Event-trace generation
  • Automatic diagnose of inefficient runtime behavior

Cons:

  • "It does not work very well with long running programs" -Kojak's authors



PGI-CDK

http://www.pgroup.com/products/cdkindex.htm

Pros:

  • MPI+OpenMP Profiling
  • Includes the PGI Fortran Compiler and other cluster tools
  • Powerful?
  • Good documentation and support
  • Measure scalability between multiple execution runs with varying number of processes/threads [4]
  • Supports both sample-based and intsrumentation based profiling

Cons:

  • Commercial, price on chart
  • Annual Service Subscription


 

Opt

http://www.allinea.com/index.php?page=74

Pros:

  • MPI+OpenMP Profiling
  • Cross-platform
  • OPT can compare multiple runs to assess code scalability [3]
  • No need to instrument your code
  • View statistical overviews and detailed views of user-defined metrics: for multiple jobs, across chosen processors and against time [3]
  • Configure function measurement and granularity at run time [3]
  • Grid-enabled

Cons:

  • Commercial Software
  • Price currently unknown



MPE/Jumpshot

http://www-unix.mcs.anl.gov/perfvis/software/viewers/index.htm

Pros:

  • MPI Logging/tracing
  • Cross-Platform (just need java2 runtime)
  • Distributed with MPICH
  • Scales well
  • Free
  • Auto-instrumentation

Cons:

  • The auto-instrumentation could use some work
  • Only event-based metrics
  • Doesn't profile OpenMP
  • Mostly only deals with MPI


MPIP

http://mpip.sourceforge.net/

Pros:

  • Good high level statistics about MPI calls
  • Scales Well
  • Free!
  • Might be useful in the beginning

Cons:

  • Only does MPI, no OpenMP
  • No call graphs
  • Probably not detailed enough once we really get into profiling


Paradyn

http://www.paradyn.org/

Pros:

  • Dynamic binary instrumentation - no need to modify code
  • Free
  • Scales extremely well
  • MPI Profiling
  • Can add user defined metrics
  • Good documentation

Cons:

  • No OpenMP Profiling
  • Buggy
  • Steep learning curve


svPablo

http://www.renci.org/software/svpablo/

Pros:

  • Supports MPI, OpenMP
  • GUI based interactive instrumentation
  • Really Good documentation
  • Free
  • Scalability analysis

Cons:

  • Buggy
  • Decent learning curve
  • No communication profiling


VTune

http://www.intel.com/cd/software/products/asmo-na/eng/vtune/vlin/239145.htm

Pros:

  • Powerful?
  • Supports MPI, OpenMP
  • Scales very well
  • Modification of code not necessary
  • Programming Language and Compiler Independent

Cons:

  • Only for Intel platforms
  • $280 per academic license

[1] http://www.llnl.gov/computing/tutorials/performance_tools/HighPerformanceToolsTechnologiesLC.pdf
[2] http://cache-www.intel.com/cd/00/00/32/37/323761_323761.pdf
[3] http://www.allinea.com/OPT.pdf
[4]
http://www.pgroup.com/products/cdkindex.htm