< previous page	page_78	next page >
Page 78 Figure 2.14 Where to put the cycle boundary. Figure 2.13): sDt ³ T + sC sDt - T = s(tm + C) - T = stm + sC - T, where tm is simply Dt - C, the time between clocks. There are two distinct overheads here; sC is the clocking overhead and it grows as s grows. The other overhead is: or the quantization overhead. Presumably, if tm were made small and s correspondingly large, then sitm - ti® 0 for all i, where si is the smallest integer to make the above equal to or greater than zero. The overall effect of quantization is stm - T. In reducing tm to reduce quantization overhead, we increase s and hence increase the clocking overhead. There is a delicate balance between the two overheads, as illustrated in the following study. Study 2.2 Cycle partitioning Suppose a processor's instruction execution requires a sequence of five events (Figure 2.14) A, B, C, D, and E, with respective delays of: A 12 ns B 24ns C 15 ns D 24 ns E 19 ns
< previous page	page_78	next page >