eugen.betke
/
ddn-ime-evaluation
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
Browse Source

Start paper writing

master
Eugen Betke 4 years ago
parent
e6092144c0
commit
d7cfc9301f
3 changed files with 115 additions and 12 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 20
      benchmark/eval_analysis_v2.R
  2. BIN
      benchmark/results.db
  3. 107
      paper/paper.tex

20
benchmark/eval_analysis_v2.R
View File

@ -58,14 +58,14 @@ dbdata$lab_count <- sprintf('Iteration %d', dbdata$count)
max_y=3500
for (lab_fs in unique(dbdata$lab_fs)) {
#for (scale in c("linear", "logarithmic")) {
for (scale in c("linear")) {
for (scale in c("linear", "logarithmic")) {
#for (scale in c("linear")) {
p = ggplot(data=dbdata[dbdata$lab_fs==lab_fs,], aes(x=nn, y=bwMiB, colour=as.factor(blocksize/1024), group=blocksize), ymin=0) +
#aes(x=nn, y=bwMiB) +
ggtitle(sprintf("%s independent random access to a shared file with IOR", lab_fs)) +
facet_grid(ppn ~ lab_access + lab_count, labeller = labeller(nn = as_labeller(nn_lab), ppn = as_labeller(ppn_lab))) +
#facet_grid(ppn ~ lab_access, labeller = labeller(nn = as_labeller(nn_lab), ppn = as_labeller(ppn_lab))) +
#facet_grid(ppn ~ lab_access + lab_count, labeller = labeller(nn = as_labeller(nn_lab), ppn = as_labeller(ppn_lab))) +
facet_grid(ppn ~ lab_access, labeller = labeller(nn = as_labeller(nn_lab), ppn = as_labeller(ppn_lab))) +
xlab("Nodes") +
ylab("Performance in MiB/s") +
theme(axis.text.x=element_text(angle=90, hjust=0.95, vjust=0.5)) +
@ -87,8 +87,8 @@ for (scale in c("linear")) {
filename_eps = sprintf("%s/%s_%s_performance.eps", folder_out, to_fn_str(lab_fs), scale)
filename_png = sprintf("%s/%s_%s_performance.png", folder_out, to_fn_str(lab_fs), scale)
ggsave(filename_png, width = 16, height = 8)
ggsave(filename_eps, width = 16, height = 8)
ggsave(filename_png, width = 8, height = 8)
ggsave(filename_eps, width = 8, height = 8)
#system(sprintf("epstopdf %s", filename_eps))
system(sprintf("rm %s", filename_eps))
@ -96,8 +96,8 @@ for (scale in c("linear")) {
aes(x=nn, y=(PortXmitData + PortRcvData) * 4 / ppn, colour=as.factor(blocksize/1024), group=blocksize) +
#aes(x=nn, y=bwMiB) +
ggtitle(sprintf('%s Infiniband throughput (PortRcvData and PortXmitData by "perfquery -x")', lab_fs)) +
facet_grid(ppn ~ lab_access + lab_count, labeller = labeller(nn = as_labeller(nn_lab), ppn = as_labeller(ppn_lab))) +
#facet_grid(ppn ~ lab_access, labeller = labeller(nn = as_labeller(nn_lab), ppn = as_labeller(ppn_lab))) +
#facet_grid(ppn ~ lab_access + lab_count, labeller = labeller(nn = as_labeller(nn_lab), ppn = as_labeller(ppn_lab))) +
facet_grid(ppn ~ lab_access, labeller = labeller(nn = as_labeller(nn_lab), ppn = as_labeller(ppn_lab))) +
xlab("Nodes") +
ylab("Performance in MiB/s") +
theme(axis.text.x=element_text(angle=90, hjust=0.95, vjust=0.5)) +
@ -116,8 +116,8 @@ for (scale in c("linear")) {
filename_eps = sprintf("%s/%s_%s_ib.eps", folder_out, to_fn_str(lab_fs), scale)
filename_png = sprintf("%s/%s_%s_ib.png", folder_out, to_fn_str(lab_fs), scale)
ggsave(filename_png, width = 16, height = 8)
ggsave(filename_eps, width = 16, height = 8)
ggsave(filename_png, width = 8, height = 8)
ggsave(filename_eps, width = 8, height = 8)
#system(sprintf("epstopdf %s", filename_eps))
system(sprintf("rm %s", filename_eps))

BIN
benchmark/results.db
View File

Binary file not shown.

107
paper/paper.tex
View File

@ -38,7 +38,7 @@
\usepackage{float}
\graphicspath{
{./fig/}
{../benchmark/eval_analysis_output/}
{../benchmark/eval_analysis_output_v2/}
}
%\newcolumntype{P}[1]{>{\raggedright\let\newline\\\arraybackslash\hspace{0pt}}m{#1\textwidth}}
@ -63,7 +63,56 @@ University of Reading, Reading, United Kingdom,\\
\keywords{Lustre, FUSE, evaluation, flash-based storage}
\section{DDN-IME}
The IOR stonewalling option (-d 60) was used together with the wear out option (-o StonewallingWearOut=1).
In this way we make sure, that each node writes the same amount of data, even if the stonewalling deadline is achieved.
For read access a 12TB test file was pre-created and used for all experiments.
Since IME has the ability to migrate data from back-end file system to IMEs and vice versa, we made sure that this file was always locate on IMEs.
For write access we used a separate test file, which was always deleted after each test run.
\begin{table}
\centering
\begin{tabular}{l|l}
Parameter & Values \\
\hline
Number of nodes & 1, 2, 4, 8, 16 \\
Processes per node & 1, 4, 8, 16 \\
Access & read, write \\
Block sizes & 16KiB, 100KiB, 1MiB, 10MiB \\
Iterations & 3 \\
\end{tabular}
\caption{Explored parameteres}
\label{tab:params:ime}
\end{table}
During the parameter exploration 540 experimentes were conducted.
The explored parameters are showed in \Cref{tab:params:ime}.
The resulting peformance is showed in \Cref{fig:perf:linear:ime}.
\begin{figure}
\begin{subfigure}{.5\textwidth}
\centering
\includegraphics[width=0.9\textwidth]{ime_linear_performance.png}
\caption{Linear scale}
\label{fig:perf:linear:ime}
\end{subfigure}
\begin{subfigure}{.5\textwidth}
\centering
\includegraphics[width=0.9\textwidth]{ime_logarithmic_performance.png}
\caption{Logarithmic scale}
\label{fig:perf:log:ime}
\end{subfigure}
\caption{Native IME random performance to a shared file measured with IOR}
\end{figure}
\textbf{Reproducible performance}
\textbf{Good scaling behaviour for large blocks}
\textbf{Limited scaling behavour for small block sizes}
\section{IBM Spectrum Scale}
% https://support.dvsus.com/hc/en-us/articles/212928566-GPFS-general-commands-
@ -74,6 +123,60 @@ University of Reading, Reading, United Kingdom,\\
A GPFS 4.2.3.0 file system was used as back-end for IMEs.
The pagepool was 6GB large.
\begin{table}
\centering
\begin{tabular}{l|l}
Parameter & Values \\
\hline
Number of nodes & 1, 2, 4, 8, 16 \\
Processes per node & 1, 4, 8 \\
Access & read, write \\
Block sizes & 16KiB, 100KiB, 1MiB, 10MiB \\
Iterations & 3 \\
\end{tabular}
\caption{Explored parameteres}
\label{tab:params:gpfs}
\end{table}
During the parameter exploration 540 experimentes were conducted.
The explored parameters are showed in \Cref{tab:params:ime}.
The resulting peformance is showed in \Cref{fig:perf:linear:ime}.
\begin{figure}
\begin{subfigure}{.5\textwidth}
\centering
\includegraphics[width=0.8\textwidth]{ibm_spectrum_scale_linear_performance.png}
\caption{Linear scale}
\label{fig:perf:linear:gpfs}
\end{subfigure}
\begin{subfigure}{.5\textwidth}
\centering
\includegraphics[width=0.8\textwidth]{ibm_spectrum_scale_logarithmic_performance.png}
\caption{Logarithmic scale}
\label{fig:perf:log:gpfs}
\end{subfigure}
\caption{Native IME random performance to a shared file measured with IOR}
\label{fig:perf:gpfs}
\end{figure}
\begin{figure}
\begin{subfigure}{.5\textwidth}
\centering
\includegraphics[width=0.8\textwidth]{ibm_spectrum_scale_linear_ib.png}
\caption{Linear scale}
\label{fig:ib:linear:gpfs}
\end{subfigure}
\begin{subfigure}{.5\textwidth}
\centering
\includegraphics[width=0.8\textwidth]{ibm_spectrum_scale_logarithmic_ib.png}
\caption{Logarithmic scale}
\label{fig:ib:log:gpfs}
\end{subfigure}
\caption{Native IME random performance to a shared file measured with IOR}
\label{fig:ib:gpfs}
\end{figure}
\subsection{Benchmarks}

Write Preview
Loading…
Cancel
Save