EnzoV1Performance

From Point
Revision as of 18:25, 2 June 2009 by Scottb (talk | contribs) (New page: =ENZO version 1 performance results= This is a short overview of the performance result from the ENZO application. For each experiment we used these inits/param files: * [http://giusto.ni...)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

ENZO version 1 performance results

This is a short overview of the performance result from the ENZO application. For each experiment we used these inits/param files:

This is a relatively small experiment but was sufficient to generate some interesting performance results. For this study we used the TAU Performance System® to gather information about ENZO's performance, in particular we are interested in the performance of the AMR simulation at scale. We ran these experiments on NCSA's Intel 64 Linux Cluster (Abe).

TAU Measurement overhead

Here is a short table listing the run-times for various experiments and the instrumentation overhead observed. Each run was on 64 processors (8 nodes).

Run Type Runtime (seconds) Overhead %
Uninstrumented runtime 1072 NA
Trace of only MPI event 1085 4.8%
Profile of all significant events 1136 6.0%
Profile with Call-path information 1196 11.6%
Profile of each Phase of execution 1208 12.7%

Runtime Breakdown on 64 processors

Here is a chart showing the contribution each function makes to the overall runtime. Notice that MPI communication time takes over 60% of the total runtime.

MeanFunctionLinux.png

Experiment Scalability

These chart show the relative efficiency for a grid size of 128^3 and 256^3. Relative efficiency is the measure of how far an run of the application is slower compared to ideal efficiency. In this case, ideal efficiency would mean a doubling of the processor count would reduces the runtime in half.

Scaling128.png Scaling256.png


This is chart show the breakdown in the runtime of different functions across different numbers of processors (128^3 grid size). MPI communication time, like in the 64 processor case, continues to dominate the runtime--and to an even greater extent when a larger number of processors are involved.

MeanRuntineAtScale2.png

Experiment Trace

This graphic shows how load imbalances causes long wait times for MPI_Allreduce. Some processors are experiencing as much as 8 seconds of wait time per reduce.

Trace.png

Experiment Call-Paths

We observe the follow relationships in the experiment call-path:

  • Almost all the time spend in MPI_Bcast is when it is called from MPI_Allreduce.
  • Almost all the time spend in MPI_Recv is when it is called from grid::CommunicationSendRegion.
  • Most all the time spend in MPI_Allgather is when it is called from CommunicationShareGrids.
  • Almost all the time spend in MPI_Allreduce is when it is called from CommunicationMinValue.

This chart show the details:

CallpathRuntime3.png

Experiment Phases

We also looked at ENZO's runtime through each iteration on the main loop in EvolveHierarchy. Here is the CommunicationShareGrids function representing the computation work done during the consecutive loops (time is microseconds)

CommunicationShareGrids2.png

Notice that some iterations are involved in writing out the grid (lots of time spent in WriteDataHierarchy).

Here is a breakdown, by function, of the time spent over the course of the experiment. Y-axis is exclusive time spend in each function, and X-axis is overall elapsed runtime:

Snapshot.png