|
|
|
|
|
|
|
Figure 6.11
Processor equivalence. |
|
|
|
|
|
|
|
|
One of the best known memory models is that of Hellerman [125]. Hellerman assumes a single sequence of address. Addresses are examined until a match between two addresses occurs in the w low-order bit positions (w = log2 m; m = the number of memory modules). The average length of independent (unmatched) addresses (uniform access pattern) then determines the bandwidth. No referencing is allowed after a match is found, since the modeling assumption is that no address queue is present and no out-of-order requests are possible (Figure 6.12). Under these conditions, the maximum available bandwidth is found to be approximately: |
|
|
|
|
|
|
|
|
The lack of queueing facilities limits the applicability of this model to simple unbuffered processors with strict in-order referencing to memory. |
|
|
|
|
|
|
|
|
The achieved bandwidth is a function of the offered request rate,ls (Figure 6.13): |
|
|
|
|
|
|
|
|
This model [273] (later developed independently by Ravi [243]) assumes that there are n simple processor requests made per memory cycle and |
|
|
|
|
|
