Benchmarks

 

This appendix provides benchmarks of SDPpack Version 0.9 BETA on some randomly generated test problems, a set of 16 LMI problems from control applications (some of which apparently are infeasible), and a set of 8 problems from truss topology design. The control and truss design problems are available as mat files from the SDPpack home page.

All problems were solved using sql.m. The random problems and the truss design problems used the default option values seet by setopt.m, i.e.:

• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 

The control LMI problems were solved with the same parameter settings above, but with . The benchmarks were conducted on a dual-processor Sparc Ultra-2 with 192 MB of main memory.

Table 3 shows a set of random problems with semidefinite blocks only. Each problem has three semidefinite blocks of equal size. The number n shown in the table is the sum of the block sizes. Thus, for example, the largest problem shown has three blocks each of size 100.

Table 4 shows a set of random problems with quadratic blocks only. Each problem has three quadratic blocks of equal size. The number n shown in the table is the sum of the block sizes. Thus, for example, the largest problem shown has three blocks each of size 250.

Tables 5 and 6 show the truss and control LMI results respectively. When the symbol * appears next to the value of termflag, this indicates that a restart was necessary.

  
Table 3: Randomly generated with semidefinite blocks only

  
Table 4: Randomly generated with quadratic blocks only

  
Table 5: Benchmarks on problems from truss topology design

  
Table 6: Benchmarks on LMI problems from control applications