source: docs/HPCA2012/06-scalability.tex @ 1335

Last change on this file since 1335 was 1335, checked in by ashriram, 8 years ago

Working on evaluation. Fixed Figure sizes

File size: 3.0 KB
Line 
1\section{Scalability}
2\subsection{Performance}
3Figure \ref{Scalability} (a) demonstrates the average XML
4well-formedness checking performance of Parabix2 for each of the
5workloads and as executed on each of the processor cores --- \CO\,
6\CITHREE\ and \SB{}.  Processing time is shown in terms of bit stream
7based operations executed in `bit-space' and postprocessing operations
8executed in `byte-space'.  In the Parabix2 parser, bit-space parallel
9bit stream parser operations consist primarily of SIMD instructions;
10byte-space operations consist of byte comparisons across arrays of
11values. Executing Parabix2 on \CITHREE{} over \CO\ results in an
12average performance improvement of 17\% in bit stream processing
13whereas migrating Parabix2 from \CITHREE{} to \SB{} results in a 22\%
14average performance gain. Bit space measurements are stable and
15consistent across each of the source inputs and cores. Postprocessing
16operations demonstrate data dependent variance. Performance gains from
1718\% to 31\% performance are observered in migrating Parabix2 from
18\CO\ to \CITHREE{}; 0\% to 17\% performance from \CITHREE\ to
19\SB{}. For the purpose of comparison, Figure \ref{Scalability} (b)
20shows the performance of the Expat parser on each of the processor
21cores.  A performance improvement of less than 5\% is observed when
22executing Expat on \CITHREE\ over \CO\ and less than 10\% on \SB\ over
23\CITHREE{}.
24
25Overall, Parabix2 scales better than Expat. Simply executing identical
26Parabix2 object code on \SB\ results in an overall performance
27improvement up to 26\%. Additional performance aspects of Parabix2 on
28\SB\ with AVX instructions are discussed in the following sections.
29
30\begin{figure}
31\centering
32\subfigure[Parabix2]{
33\includegraphics[width=0.40\textwidth]{plots/P2_scalability.pdf}
34}
35\subfigure[Expat]{
36\includegraphics[width=0.40\textwidth]{plots/Expat_scalability.pdf}
37}
38\caption{Average Performance Parabix vs. Expat (y-axis: CPU Cycles per kB)}
39\label{Scalability}
40\end{figure}
41
42
43\subsection{Power and Energy}
44
45Figure \ref{power_Parabix2} shows the average power consumption of
46Parabix2 over each workload and as executed on each of the processor
47cores --- \CO{}, \CITHREE\ and \SB{}.  Average power consumption on
48\CO{} is 32 watts. Execution on \CITHREE\ results in 30\% power saving
49over \CO{}\SB\ saves 25\% of the power compared with \CITHREE\ and
50consumes only 15 watts.
51
52In XML parsing we observe energy consumption is dependent on processing time. That is, a reduction in processing time results in a directly proportional reduction in energy consumption.
53With newer processor cores comes improvements in application performance. As a result, Parabix2 executed on \SB\ consumes 72\% to 75\% less energy than Parabix2 on \CO{}.
54
55
56
57
58\begin{figure}
59\centering
60\subfigure[Avg. Power of Parabix on various hardware (Watts)]{
61\includegraphics[width=85mm]{plots/power_Parabix2.pdf}
62\label{power_Parabix2}
63}
64\hfill
65\centering
66\subfigure[Avg. Energy Consumption on various hardware (nJ per kB)]{
67\includegraphics[width=85mm]{plots/energy_Parabix2.pdf}
68\label{energy_Parabix2}
69}
70\end{figure}
Note: See TracBrowser for help on using the repository browser.