Congestion Balancing Global Router - VLSI Design and Test - page 230

Tabl e 1. The results of five state-of-the-art global routers and CSaCB on ISPD'98
benchmark. The runtimes for each router are given in seconds.

Circuit BoxRouter 2.0 FastRoute 2 .0 FGR
TOF WL CT TOF WL CT TOF WL CT
ibm01 0 62659 32.8 31 68489 0.72 0 63332 10
ibm02 0 171110 35.9 0 178868 0.93 0 168918 13
ibm03 0 146634 17.6 0 150393 0.6 0 146412 5
ibm04 0 167275 115.9 64 175037 1.88 0 167101 29
ibm06 0 277913 47.4 0 284935 1.36 0 277608 18
ibm07 0 365790 85.9 0 375185 1.6 0 366180 20
ibm08 0 405634 90.1 0 411703 2.36 0 404714 18
ibm09 0 413862 273.1 3 424949 1.92 0 413053 20
ibm10 0 590141 352.4 0 595622 2.79 0 578795 92
Circuit NTHU-Route 2.0 MaizeRouter CSaCB
TOF WL CT TOF WL CT TOF WL CT

ibm01 0 62498 2.4 0 63720 NA 58 65012 2.6

ibm02 0 169881 3.3 0 170342 NA 36 170132 5.3

ibm03 0 146458 2.5 0 147078 NA 21 148338 2.7

ibm04 0 166452 5.9 0 170095 NA 122 170556 6.2

ibm06 0 277696 5.5 0 279566 NA 42 281124 6.2

ibm07 0 366133 6.4 0 369340 NA 84 366890 6.2

ibm08 0 404976 5.9 0 406349 NA 40 405684 6.8

ibm09 0 414738 5.7 0 415852 NA 18 416214 7.4

ibm10 0 579870 12.3 0 585921 NA 51 585654 15.3

Tabl e 2. List machines used by other routers

Router CPU Memory

BoxRouter 2.0 Intel Pentium 4 2.8Ghz 2GB

FastRoute 2.0 Intel Pentium 4 3.0 GHz 2GB

FGR AMD Opteron 2.4 Ghz 4GB

NTHU-Route 2.0 AMD Opteron 2.2Ghz 8GB

MaizeRouter Dual-core 2.8-GHz AMD Opteron 16 GB

To empirically evaluate the performance of our Congestion Sensing and Con-
gestion Balancing (CSaCB) technique, we compare it to six recent popular state-
of-the-art academic global routers: FastRoute 2.0 [18], BoxRouter [23], FGR [16],
NTHU-Route 2.0 [13], and MaizeRouter [20]. We compare the results in terms
of total overflow, wirelength, and CPU time. The column heads “TOF”, “WL”,
and “CT” denote the total overflow, total wire length, and CPU time required, in
Table 1. All the existing router achieve 0 overflow for all nine circuits except Fas-
tRoute 2.0 which achieve 0 overflow for six out of nine circuits. The wirelength
of NTHU-Route 2.0 is the least among all the routers. Our proposed technique
CSaCB also achieves fairly good results with some overflow. Being an ongoing
work, CSaCB requires few improvements and we hope the proposed technique
will work well if we are able to incorporate these improvements. Table 2 shows
the list of machines used by each router.