Remove redundant \usepackage.

[mpi-energy.git] / paper.tex

diff --git a/paper.tex b/paper.tex
index 9c374ba8c8b4d56d6ca1346ca15f2942f2f41b28..e62908a6ffab9c5c2b8a14a1e0f6de133fba4da7 100644 (file)
--- a/paper.tex
+++ b/paper.tex
@@ -4,7 +4,7 @@
 \usepackage[utf8]{inputenc}
 \usepackage[english]{babel}
 \usepackage{algpseudocode}
 \usepackage[utf8]{inputenc}
 \usepackage[english]{babel}
 \usepackage{algpseudocode}

-\usepackage{graphicx,graphics}
+\usepackage{graphicx}

 \usepackage{subfig}
 \usepackage{amsmath}
 
 \usepackage{subfig}
 \usepackage{amsmath}
 


@@ -40,6 +40,7 @@
 \newcommand{\Pdyn}{\Xsub{P}{dyn}}
 \newcommand{\PnormInv}{\Xsub{P}{NormInv}}
 \newcommand{\Pnorm}{\Xsub{P}{Norm}}
 \newcommand{\Pdyn}{\Xsub{P}{dyn}}
 \newcommand{\PnormInv}{\Xsub{P}{NormInv}}
 \newcommand{\Pnorm}{\Xsub{P}{Norm}}

+\newcommand{\Tnorm}{\Xsub{T}{Norm}}

 \newcommand{\Pstates}{\Xsub{P}{states}}
 \newcommand{\Pstatic}{\Xsub{P}{static}}
 \newcommand{\Sopt}{\Xsub{S}{opt}}
 \newcommand{\Pstates}{\Xsub{P}{states}}
 \newcommand{\Pstatic}{\Xsub{P}{static}}
 \newcommand{\Sopt}{\Xsub{S}{opt}}


@@ -136,7 +137,7 @@ the MPI program to choose the scaling factor.  This algorithm has the ability to
 predict both energy consumption and execution time over all available scaling
 factors.  The prediction achieved depends on some computing time information,
 gathered at the beginning of the runtime.  We apply this algorithm to the NAS parallel benchmarks (NPB v3.3)~\cite{44}.  Our experiments are executed using the simulator
 predict both energy consumption and execution time over all available scaling
 factors.  The prediction achieved depends on some computing time information,
 gathered at the beginning of the runtime.  We apply this algorithm to the NAS parallel benchmarks (NPB v3.3)~\cite{44}.  Our experiments are executed using the simulator

-SimGrid/SMPI v3.10~\cite{Casanova:2008:SGF:1397760.1398183} over an homogeneous
+SimGrid/SMPI v3.10~\cite{Casanova:2008:SGF:1397760.1398183} over a homogeneous

 distributed memory architecture.  Furthermore, we compare the proposed algorithm
 with Rauber and Rünger methods~\cite{3}.  The comparison's results show that our
 algorithm gives better energy-time trade-off.
 distributed memory architecture.  Furthermore, we compare the proposed algorithm
 with Rauber and Rünger methods~\cite{3}.  The comparison's results show that our
 algorithm gives better energy-time trade-off.


@@ -228,7 +229,7 @@ our paper is to present a new online scaling factor selection method which has t
 %   paper in homogeneous clusters}
 
 
 %   paper in homogeneous clusters}
 
 

-\section{Energy model for homogeneous platform}
+\section{Energy model for a homogeneous platform}

 \label{sec.exe}
 Many researchers~\cite{9,3,15,26} divide the power consumed by a processor into
 two power metrics: the static and the dynamic power.  While the first one is
 \label{sec.exe}
 Many researchers~\cite{9,3,15,26} divide the power consumed by a processor into
 two power metrics: the static and the dynamic power.  While the first one is


@@ -285,11 +286,11 @@ function of the scaling factor $S$, as in EQ~\eqref{eq:energy}.
     \left( T_1 + \sum_{i=2}^{N} \frac{T_i^3}{T_1^2} \right) +
       \Pstatic \cdot T_1 \cdot S_1 \cdot N
 \end{equation}
     \left( T_1 + \sum_{i=2}^{N} \frac{T_i^3}{T_1^2} \right) +
       \Pstatic \cdot T_1 \cdot S_1 \cdot N
 \end{equation}

-where $N$ is the number of parallel nodes, $T_i$ and $S_i$ for $i=1,\dots,N$ are
-the execution times and scaling factors of the sorted tasks.  Therefore, $T_1$ is
+where $N$ is the number of parallel nodes, $T_i$ for $i=1,\dots,N$ are
+the execution times of the sorted tasks.  Therefore, $T_1$ is

 the time of the slowest task, and $S_1$ its scaling factor which should be the
 highest because they are proportional to the time values $T_i$.  The scaling
 the time of the slowest task, and $S_1$ its scaling factor which should be the
 highest because they are proportional to the time values $T_i$.  The scaling

-factors are computed as in EQ~\eqref{eq:si}.
+factors $S_i$ are computed as in EQ~\eqref{eq:si}.

 \begin{equation}
   \label{eq:si}
   S_i = S \cdot \frac{T_1}{T_i}
 \begin{equation}
   \label{eq:si}
   S_i = S \cdot \frac{T_1}{T_i}


@@ -310,8 +311,8 @@ EQ~\eqref{eq:sopt}.
 \section{Performance evaluation of MPI programs}
 \label{sec.mpip}
 
 \section{Performance evaluation of MPI programs}
 \label{sec.mpip}
 

-The performance (execution time) of parallel synchronous MPI applications depends
-on the time of the slowest task.  If there is no
+The execution time of a parallel synchronous iterative application is
+equal to the execution time of the slowest task.  If there is no

 communication and the application is not data bounded, the execution time of a
 parallel program is linearly proportional to the operational frequency and any
 DVFS operation for energy reduction increases the execution time of the parallel
 communication and the application is not data bounded, the execution time of a
 parallel program is linearly proportional to the operational frequency and any
 DVFS operation for energy reduction increases the execution time of the parallel


@@ -336,48 +337,34 @@ for each processor as presented in the next section.
 \section{Performance and energy reduction trade-off}
 \label{sec.compet}
 
 \section{Performance and energy reduction trade-off}
 \label{sec.compet}
 

-This section presents our approach for choosing the optimal scaling factor.
-This factor gives maximum energy reduction while taking into account the execution
-times for both computation and communication.  The relation between the performance
-and the energy is nonlinear and complex. Thus, unlike the relation between the performance and  the scaling factor,  the relation of energy with the scaling factor is nonlinear, for more details refer to~\cite{17}.  Moreover, they are not measured using the same metric.  To
-solve this problem, we normalize the energy by calculating the ratio between
-the consumed energy with scaled frequency and the consumed energy without scaled
+This section presents our method for choosing the optimal scaling factor that
+gives the best tradeoff between energy reduction and performance. This method
+takes into account the execution times for both computation and communication to
+compute the scaling factor.  Since the energy consumption and the performance
+are not measured using the same metric, a normalized value of both measurements
+can be used to compare them.  The normalized energy is the ratio between the
+consumed energy with scaled frequency and the consumed energy without scaled

 frequency:
 \begin{multline}
   \label{eq:enorm}
 frequency:
 \begin{multline}
   \label{eq:enorm}

-  \Enorm(S) = \frac{ \Ereduced}{\Eoriginal} \\
-         {} = \frac{\Pdyn \cdot S^{-2} \cdot
+  \Enorm = \frac{ \Ereduced}{\Eoriginal} \\
+      {} = \frac{\Pdyn \cdot S_1^{-2} \cdot

              \left( T_1 + \sum_{i=2}^{N}\frac{T_i^3}{T_1^2}\right) +
              \left( T_1 + \sum_{i=2}^{N}\frac{T_i^3}{T_1^2}\right) +

-               \Pstatic \cdot T_1 \cdot S \cdot N}{
+               \Pstatic \cdot T_1 \cdot S_1 \cdot N}{

              \Pdyn \cdot \left(T_1+\sum_{i=2}^{N}\frac{T_i^3}{T_1^2}\right) +
                \Pstatic \cdot T_1 \cdot N }
 \end{multline}
              \Pdyn \cdot \left(T_1+\sum_{i=2}^{N}\frac{T_i^3}{T_1^2}\right) +
                \Pstatic \cdot T_1 \cdot N }
 \end{multline}

-In the same way we can normalize the performance as follows:
+In the same way, the normalized execution time of a program is computed as follows:

 \begin{equation}
   \label{eq:pnorm}
 \begin{equation}
   \label{eq:pnorm}

-  \Pnorm(S) = \frac{\Tnew}{\Told}
-            = \frac{\TmaxCompOld \cdot S + \TmaxCommOld}{
+  \Tnorm = \frac{\Tnew}{\Told}
+         = \frac{\TmaxCompOld \cdot S + \TmaxCommOld}{

              \TmaxCompOld + \TmaxCommOld}
 \end{equation}
              \TmaxCompOld + \TmaxCommOld}
 \end{equation}

-The second problem is that the optimization operation for both energy and
-performance is not in the same direction.  In other words, the normalized energy
-and the performance curves are not at the same direction see
-Figure~\ref{fig:rel}\subref{fig:r2}.  While the main goal is to optimize the
-energy and performance in the same time.  According to the
-equations~\eqref{eq:enorm} and~\eqref{eq:pnorm}, the scaling factor $S$ reduce
-both the energy and the performance simultaneously.  But the main objective is
-to produce maximum energy reduction with minimum performance reduction.  Many
-researchers used different strategies to solve this nonlinear problem for
-example see~\cite{19,42}, their methods add big overheads to the algorithm to
-select the suitable frequency.  In this paper we present a method to find the
-optimal scaling factor $S$ to optimize both energy and performance
-simultaneously without adding a big overhead.  Our solution for this problem is
-to make the optimization process for energy and performance follow the same
-direction.  Therefore, we inverse the equation of the normalized performance as
-follows:
+The relation between the execution time and the consumed energy of a program is nonlinear and complex. In consequences,  the relation between the consumed energy and the scaling factor is also nonlinear, for more details refer to~\cite{17}. Therefore, the resulting normalized energy consumption curve and execution time curve, for different scaling factors, do not have the same direction see Figure~\ref{fig:rel}\subref{fig:r2}.  To tackle this problem and optimize both terms,  we inverse the equation of the normalized execution time as follows:

 \begin{equation}
   \label{eq:pnorm_en}
 \begin{equation}
   \label{eq:pnorm_en}

-  \Pnorm^{-1}(S) = \frac{ \Told}{ \Tnew}
+  \Pnorm = \frac{ \Told}{ \Tnew}

                = \frac{\TmaxCompOld +
                  \TmaxCommOld}{\TmaxCompOld \cdot S +
                  \TmaxCommOld}
                = \frac{\TmaxCompOld +
                  \TmaxCommOld}{\TmaxCompOld \cdot S +
                  \TmaxCommOld}


@@ -392,7 +379,7 @@ follows:
   \label{fig:rel}
 \end{figure}
 Then, we can model our objective function as finding the maximum distance
   \label{fig:rel}
 \end{figure}
 Then, we can model our objective function as finding the maximum distance

-between the energy curve EQ~\eqref{eq:enorm} and the inverse of performance
+between the energy curve EQ~\eqref{eq:enorm} and the inverse of the execution time (performance)

 curve EQ~\eqref{eq:pnorm_en} over all available scaling factors.  This
 represents the minimum energy consumption with minimum execution time (better
 performance) at the same time, see Figure~\ref{fig:rel}\subref{fig:r1}.  Then
 curve EQ~\eqref{eq:pnorm_en} over all available scaling factors.  This
 represents the minimum energy consumption with minimum execution time (better
 performance) at the same time, see Figure~\ref{fig:rel}\subref{fig:r1}.  Then


@@ -400,7 +387,7 @@ our objective function has the following form:
 \begin{equation}
   \label{eq:max}
   \MaxDist = \max_{j=1,2,\dots,F}
 \begin{equation}
   \label{eq:max}
   \MaxDist = \max_{j=1,2,\dots,F}

-      (\overbrace{\Pnorm^{-1}(S_j)}^{\text{Maximize}} -
+      (\overbrace{\Pnorm(S_j)}^{\text{Maximize}} -

        \overbrace{\Enorm(S_j)}^{\text{Minimize}} )
 \end{equation}
 where $F$ is the number of available frequencies. Then we can select the optimal
        \overbrace{\Enorm(S_j)}^{\text{Minimize}} )
 \end{equation}
 where $F$ is the number of available frequencies. Then we can select the optimal


@@ -433,17 +420,20 @@ the objective function described above.
     \For {$j = 2$ to $\Pstates$}
       \State $\Fnew \gets \Fnew - \Fdiff$
       \State $S \gets \Fmax / \Fnew$
     \For {$j = 2$ to $\Pstates$}
       \State $\Fnew \gets \Fnew - \Fdiff$
       \State $S \gets \Fmax / \Fnew$

+      \State $S_i \gets S \cdot \frac{T_1}{T_i}
+                  = \frac{\Fmax}{\Fnew} \cdot \frac{T_1}{T_i}$
+             for $i=1,\dots,N$

       \State $\Enorm \gets
       \State $\Enorm \gets

-          \frac{\Pdyn \cdot S^{-2} \cdot
+          \frac{\Pdyn \cdot S_1^{-2} \cdot

                   \left( T_1 + \sum_{i=2}^{N}\frac{T_i^3}{T_1^2}\right) +
                   \left( T_1 + \sum_{i=2}^{N}\frac{T_i^3}{T_1^2}\right) +

-                  \Pstatic \cdot T_1 \cdot S \cdot N }{
+                  \Pstatic \cdot T_1 \cdot S_1 \cdot N }{

                 \Pdyn \cdot
                   \left(T_1+\sum_{i=2}^{N}\frac{T_i^3}{T_1^2}\right) +
                   \Pstatic \cdot T_1 \cdot N }$
                 \Pdyn \cdot
                   \left(T_1+\sum_{i=2}^{N}\frac{T_i^3}{T_1^2}\right) +
                   \Pstatic \cdot T_1 \cdot N }$

-      \State $\PnormInv \gets \Told / \Tnew$
-      \If{$(\PnormInv - \Enorm > \Dist)$}
+      \State $\Pnorm \gets \Told / \Tnew$
+      \If{$(\Pnorm - \Enorm > \Dist)$}

         \State $\Sopt \gets S$
         \State $\Sopt \gets S$

-        \State $\Dist \gets \PnormInv - \Enorm$
+        \State $\Dist \gets \Pnorm - \Enorm$

       \EndIf
     \EndFor
     \State  Return $\Sopt$
       \EndIf
     \EndFor
     \State  Return $\Sopt$


@@ -521,9 +511,9 @@ Each node in the cluster has 18 frequency values
 from \np[GHz]{2.5} to \np[MHz]{800} with \np[MHz]{100} difference between each
 two successive frequencies. The nodes are connected via an ethernet network with 1Gbit/s  bandwidth. 
 
 from \np[GHz]{2.5} to \np[MHz]{800} with \np[MHz]{100} difference between each
 two successive frequencies. The nodes are connected via an ethernet network with 1Gbit/s  bandwidth. 
 

-\subsection{Performance prediction verification}
+\subsection{Execution time prediction verification}

 
 

-In this section we evaluate the precision of our performance prediction method
+In this section we evaluate the precision of our execution time prediction method

 based on EQ~\eqref{eq:tnew} by applying it to the NAS benchmarks.  The NAS programs
 are executed with the class B option to compare the real execution time with
 the predicted execution time.  Each program runs offline with all available
 based on EQ~\eqref{eq:tnew} by applying it to the NAS benchmarks.  The NAS programs
 are executed with the class B option to compare the real execution time with
 the predicted execution time.  Each program runs offline with all available


@@ -563,14 +553,14 @@ and Rünger in~\cite{3}.  The results showed that the algorithm selected
 different scaling factors for each program depending on the communication
 features of the program as in the plots from Figure~\ref{fig:nas}.  These plots
 illustrate that there are different distances between the normalized energy and
 different scaling factors for each program depending on the communication
 features of the program as in the plots from Figure~\ref{fig:nas}.  These plots
 illustrate that there are different distances between the normalized energy and

-the normalized inverted performance curves, because there are different
+the normalized inverted execution time curves, because there are different

 communication features for each benchmark.  When there are little or no
 communication features for each benchmark.  When there are little or no

-communications, the inverted performance curve is very close to the energy
+communications, the inverted execution time curve is very close to the energy

 curve.  Then the distance between the two curves is very small.  This leads to
 small energy savings.  The opposite happens when there are a lot of
 communication, the distance between the two curves is big.  This leads to more
 energy savings (e.g. CG and FT), see Table~\ref{table:compareC}.  All discovered
 curve.  Then the distance between the two curves is very small.  This leads to
 small energy savings.  The opposite happens when there are a lot of
 communication, the distance between the two curves is big.  This leads to more
 energy savings (e.g. CG and FT), see Table~\ref{table:compareC}.  All discovered

-frequency scaling factors optimize both the energy and the performance
+frequency scaling factors optimize both the energy and the execution time

 simultaneously for all NAS benchmarks.  In Table~\ref{table:compareC}, we record
 all optimal scaling factors results for each benchmark running class C.  These
 scaling factors give the maximum energy saving percentage and the minimum
 simultaneously for all NAS benchmarks.  In Table~\ref{table:compareC}, we record
 all optimal scaling factors results for each benchmark running class C.  These
 scaling factors give the maximum energy saving percentage and the minimum


@@ -601,7 +591,7 @@ savings).
 
 In this section, we compare our scaling factor selection method with Rauber and
 Rünger methods~\cite{3}.  They had two scenarios, the first is to reduce energy
 
 In this section, we compare our scaling factor selection method with Rauber and
 Rünger methods~\cite{3}.  They had two scenarios, the first is to reduce energy

-to the optimal level without considering the performance as in
+to the optimal level without considering the execution time as in

 EQ~\eqref{eq:sopt}.  We refer to this scenario as $R_{E}$.  The second scenario
 is similar to the first except setting the slower task to the maximum frequency
 (when the scale $S=1$) to keep the performance from degradation as mush as
 EQ~\eqref{eq:sopt}.  We refer to this scenario as $R_{E}$.  The second scenario
 is similar to the first except setting the slower task to the maximum frequency
 (when the scale $S=1$) to keep the performance from degradation as mush as


@@ -679,12 +669,12 @@ trade-offs such as in BT and EP.
 
 In this paper, we have presented a new online scaling factor selection method
 that optimizes simultaneously the energy and performance of a distributed
 
 In this paper, we have presented a new online scaling factor selection method
 that optimizes simultaneously the energy and performance of a distributed

-application running on an homogeneous cluster.  It uses the computation and
+application running on a homogeneous cluster.  It uses the computation and

 communication times measured at the first iteration to predict energy
 communication times measured at the first iteration to predict energy

-consumption and the performance of the parallel application at every available
+consumption and the execution time of the parallel application at every available

 frequency.  Then, it selects the scaling factor that gives the best trade-off
 between energy reduction and performance which is the maximum distance between
 frequency.  Then, it selects the scaling factor that gives the best trade-off
 between energy reduction and performance which is the maximum distance between

-the energy and the inverted performance curves.  To evaluate this method, we
+the energy and the inverted execution time curves.  To evaluate this method, we

 have applied it to the NAS benchmarks and it was compared to Rauber and Rünger
 methods while being executed on the simulator SimGrid.  The results showed that
 our method, outperforms Rauber and Rünger's methods in terms of energy-performance
 have applied it to the NAS benchmarks and it was compared to Rauber and Rünger
 methods while being executed on the simulator SimGrid.  The results showed that
 our method, outperforms Rauber and Rünger's methods in terms of energy-performance


@@ -711,6 +701,7 @@ supporting his work.
 % the document is modified later
 %\IEEEtriggeratref{15}
 
 % the document is modified later
 %\IEEEtriggeratref{15}
 

+\newpage

 \bibliographystyle{IEEEtran}
 \bibliography{IEEEabrv,my_reference}
 \end{document}
 \bibliographystyle{IEEEtran}
 \bibliography{IEEEabrv,my_reference}
 \end{document}