arma-thesis

git clone https://git.igankevich.com/arma-thesis.git
Log | Files | Refs | LICENSE

commit 3c8c3927f7362fb28c083f20531558210304bb23
parent 475afd31dfa7f47ca0cea6315b9ac0da057657a3
Author: Ivan Gankevich <igankevich@ya.ru>
Date:   Fri, 27 Jan 2017 11:03:06 +0300

Translate LB evaluation p2.

Diffstat:
phd-diss.org | 13+++++++++++++
1 file changed, 13 insertions(+), 0 deletions(-)

diff --git a/phd-diss.org b/phd-diss.org @@ -2026,6 +2026,19 @@ compared on the platform with the configuration listed in table | Disk | Seagate ST3250318AS | | Disk speed (rpm) | 7200 | +In the experiment load balancing algorithm showed higher performance than OpenMP +implementation. In each application run only the size of the realisation was +varied. The figure The more the size of the realisation is the more the gap in +performance is (as shown in figure [[fig:factory-performance]]). High performance is +a result of overlap of computation phase and data output phase (as shown in +figure [[fig:factory-overlap]]). In OpenMP implementation there is no such overlap +and data output phase begins when computation is over. In contrast to this, in +the second implementation load balancing algorithm makes both phases end almost +simultaneously. So, /pipelining internally parallel sequential phases is more +efficient than their sequential execution/, and this allows balancing the load +across different compute devices. + + #+name: fig:factory-performance #+begin_src R :results output graphics :exports results :file build/factory-vs-openmp.pdf source(file.path("R", "common.R"))