20.05.2014 Views

link to my thesis

link to my thesis

link to my thesis

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

5.3. EFFICIENCY OF THE ND-SYSTEMS APPROACH 73<br />

of all s<strong>to</strong>red points required for computing the requested point (i.e., the size of the<br />

path), the <strong>to</strong>tal number of floating point operations (flops) and the actual running<br />

time (in ms. obtained on a PC platform with an Intel Pentium PIV 2.8GHz processor<br />

and 512MB internal memory) are given for each requested point.<br />

Table 5.1: Two-dimensional case (2 × 2 initial state w 0,0 )<br />

Requested point Linear Diagonal Equalizing Axis<br />

(0,10) S<strong>to</strong>red points 36 91 46 34<br />

Flops 100 265 167 118<br />

Cpu time (ms) 7 6 2 2<br />

(0,50) S<strong>to</strong>red points 156 1431 546 154<br />

Flops 460 4285 2181 718<br />

Cpu time (ms) 32 23 25 8<br />

(0,100) S<strong>to</strong>red points 306 5356 1921 304<br />

Flops 910 16060 7697 1468<br />

Cpu time (ms) 16 40 55 20<br />

(10,10) S<strong>to</strong>red points 66 276 64 143<br />

Flops 209 820 239 598<br />

Cpu time (ms) 30 20 1 20<br />

(50,50) S<strong>to</strong>red points 306 5356 304 1723<br />

Flops 1009 16060 1199 7918<br />

Cpu time (ms) 40 40 10 60<br />

(100,100) S<strong>to</strong>red points 606 20706 604 5948<br />

Flops 2009 62110 2399 27758<br />

Cpu time (ms) 20 120 50 220<br />

Table 5.2: Three-dimensional case (2 × 2 × 2 initial state w 0,0,0 )<br />

Requested point Linear Diagonal Equalizing Axis<br />

(0,0,10) S<strong>to</strong>red points 120 560 159 88<br />

Flops 344 2216 832 399<br />

Cpu time (ms) 12 18 4 1<br />

(0,0,50) S<strong>to</strong>red points 520 27720 4825 408<br />

Flops 1544 110856 32328 2199<br />

Cpu time (ms) 23 193 89 18<br />

(0,0,100) S<strong>to</strong>red points 1020 192920 29808 808<br />

Flops 3044 771656 214696 4449<br />

Cpu time (ms) 32 1017 1875 33<br />

(10,10,10) S<strong>to</strong>red points 320 7140 248 1868<br />

Flops 1040 28536 1379 10565<br />

Cpu time (ms) 20 40 1 20<br />

(50,50,50) S<strong>to</strong>red points 1520 620620 1208 55348<br />

Flops 5040 2482456 7079 351861<br />

Cpu time (ms) 80 2774 60 2614<br />

(100,100,100) S<strong>to</strong>red points 3020 4728720 2408 281948<br />

Flops 10040 18914856 14204 1892839<br />

Cpu time (ms) 50 265008 150 781646<br />

From this experiment we may conclude the following:<br />

• The linear method indeed exhibits a linear complexity with respect <strong>to</strong> |t|. But<br />

in higher dimensions (e.g., n>10) the linear method may become inefficient<br />

because the simplex entirely covering the hypercube of initial values becomes<br />

very large. Although it constitutes a stable pattern associated with shifts in all

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!