Results of SDPLIB problems

For the description and problem data, see SDPLIB.

Selected SDPLIB problems


 no._of size_of
no._of size_of
problem var. matrix problem var. matrix
arch8 174 335 qap9 748 82
control7 136 45 qap10 1021 101
control10 1326 150 qpG11 800 1600
control11 1596 165 qpG51 1000 2000
equalG11 801 801 ss30 132 426
equalG51 1001 1001 theta3 1106 150
gpp250-4 250 250 theta4 1949 200
gpp500-4 501 500 theta5 3028 250
hinf15 91 37 theta6 4375 300
maxG11 800 800 thetaG11 2401 801
maxG32 2000 2000 thetaG51 6910 1001
maxG51 1001 1001 truss7 86 301
mcp250-1 250 250 truss8 496 628
mcp500-1 500 500



Results

problem CSDP DSDP SDPA SDPT3 SeDuMi PENNON
arch8 10 4 5 7 33 6
control7 53 114 61 48 47 82
control10 155 740 329 194 27 669
control11 218 1236 519 288 490 974
equalG11 121 139 94 156 1260 102
equalG51 320 351 261 350 3678 391
gpp250-4 11 4 3 8 56 4
gpp500-4 36 3 27 39 379 21
hinf15 1 1 1 3 2 1
maxG11 50 6 123 54 887 25
maxG32 669 72 1937 650 15084 259
maxG51 152 26 124 147 2081 172
mcp250-1 4 1 3 4 20 2
mcp500-1 16 2 19 18 239 7
qap9 2 10 3 5 19 11
qap10 4 19 6 8 44 16
qpG11 383 18 474 339 6878 68
qpG51 693 185 1003 658 22226 294
ss30 20 10 41 18 99 20
theta3 6 11 8 10 35 18
theta4 25 48 35 34 172 84
theta5 79 172 110 110 605 267
theta6 215 551 276 287 2282 797
thetaG11 127 3841 190 168 1422 394
thetaG51 14642 >20000 3041 3770 13679 9475
truss7 1 1 2 1 1 1
truss8 5 12 23 8 6 13

Test performed on Pentium IV PC (2.5 GHz) with 2GB RDRAM running Linux-2.4.19.
CPU in seconds.

To see a kind of average behaviour, in the next table we show the sum of CPU times for two sets of problems (``non-G'' and ``G''), as well as for all problems from the above table.
problem
 CSDP
 DSDP
 SDPA
 SDPT3
 SeDuMi
 PENNON
non-G 861 2936 1471 1090 4803 2993
G \ thetaG51 2515 4638 4206 2522 53516 1405
all 18018 >27574 8718 7382 71998 13873

Finally we present the performance profile of the codes on this problem collection. It shows that PENNON is not often among the fastest performers (on this set) but is indeed the most robust and efficient one.


Michal Kocvara
June 27, 2003