The static
connections are shown in Fig. 9 as dashed lines. SCF is performed
after the BS process. The scheduler first broadcasts
the BS matching result of IMs andOMs to every IM. Then, following
the static connections, each unmatched IM can send a
cell to each unmatched OM if the corresponding VOMQs do
not have a batch because these cells are still in the cell cache and
they can be read out through the northbound output ports. If
we want to forward cells from VOMQs which already have
batches, it will make “holes” in the DSM because these cells
are already distributed to the memory bank to avoid conflict.
Fortunately, the VOMQs with batches will finally be scheduled
by BS. Fig. 9 shows an example of SCF. We suppose that
IM3 andOM2 are selected by BS; therefore, all the output links
of IM3 and all the input links of OM2 are occupied; while all
the output/input links of unmatched IMs/OMs are free to
use. Following the static path, IM1 is able to send two cells,
one toOM1 and the other toOM3 through the paths labeled as
1 and 2 in Fig. 9, respectively. Similarly, IM2 also sends two
cells to OM1 and OM3 through the paths labeled as 3 and 4,
respectively. The cells are directly forwarded to the OMs
without being stored in the CMs, because the corresponding
CQs must be empty. Simulation results in Section 7 show that
SCF can reduce the latency under lowloads.