|
|
|
|
|
|
|
One of the great data integrity issues involving memory is the volatility of data in the current memory technology. As memory capacity increases, the DRAM technology gets ever closer to the cost per bit of disk, but the magnetic technology has the singular advantage of being nonvolatile. Volatile memory is easier to corrupt; hence, the need for and importance of a thoroughly checked high-integrity memory system. Breakthroughs in the area of volatility could significantly alter the whole nature of the computer system. |
|
|
|
|
|
|
|
|
Processor-memory models are well understood insofar as a single (or even multiple) processor interacts with multiple memory modules. Much less well understood is the model of the virtual address space and its correspondence to real memory. Performance across the memory hierarchy is a function of the size of the real memory, as well as many systems attributes. |
|
|
|
|
|
|
|
|
Binomial/Poisson arrivals are only an approximate description of the processor request stream to the memory system. The assumption of uniform address distribution is clearly erroneous given the sequential nature of instruction requests. The uniform address distribution assumption tends to predict a higher level of contention than actually exists in the system, i.e., provides a somewhat conservative estimate of performance. |
|
|
|
|
|
|
|
|
6.12 Annotated Bibliography |
|
|
|
|
|
|
|
|
The earliest memory contention modeling efforts were those of Flores [90] and Hellerman [125]. It is perhaps no surprise that their models represent extremes in contention modeling. With two such diametric views of processor modeling, it became obvious that the issue of contention lay in the definition of the processor-memory interaction. Thus, a good deal of attention focused on the solution of the problem of modeling n "ideal" processors, each making one request per memory cycle to m memory modules. There are many contributions to the solution of this problem, but the simplicity of the Strecker model [273], the queueing model of Baskett [29], and the comprehensive analysis of Rau [241] deserve special mention. |
|
|
|
|
|
|
|
|
The work of Chang et al. [49] represents a particularly interesting treatment of the relationship among the models and how they correspond to processor reality. |
|
|
|
 |
|
|
|
|
W. Stallings. Computer Organization and Architecture: Principles of Structure and Function. Macmillan, New York, 2nd edition, 1990. |
|
|
|
|
|
|
|
|
Intel Corp. Memory Components Handbook. Intel, Santa Clara, CA. |
|
|
|
|
|