< previous page page_421 next page >

Page 421
3. Design a SECDED (Hamming type) coding scheme for a memory with data word size of 18 bits. Show placement of each check bit and give logic equations for its action.
4. A pipelined processor has a peak performance of 40 MIPS (decoding one instruction per cycle). It creates 0.90 instruction references/instr, 0.30 data reads, and 0.10 data writes per instruction. The memory system has:
Ta
=
200ns,
Tc
=
100ns.

d87111c01013bcda00bb8640fdff6754.gif
Suppose it is possible to partition memory so that instruction references go to instruction memory and data references go to data memory. No cache is used.
d87111c01013bcda00bb8640fdff6754.gif
Using the open-queue model, compute:
d87111c01013bcda00bb8640fdff6754.gif
(a) The allocation of modules to instruction and data.
d87111c01013bcda00bb8640fdff6754.gif
(b) Effective Tw per reference (overall).
d87111c01013bcda00bb8640fdff6754.gif
(c) Queue size.
d87111c01013bcda00bb8640fdff6754.gif
(d) Comparison to a single integrated (I and D) memory system (select m, a power of 2) using the same model.
5. Suppose two processors (in a multiprocessor system) make a total of exactly two references to memory every memory cycle (Tc = 100 ns). The memory consists of eight low-order interleaved memory modules with Taccess = 120 ns. Find:
d87111c01013bcda00bb8640fdff6754.gif
(a) Expected waiting time (Tw).
d87111c01013bcda00bb8640fdff6754.gif
(b) Total access time.
d87111c01013bcda00bb8640fdff6754.gif
(c) Mean total number of queued (waiting) requests.
d87111c01013bcda00bb8640fdff6754.gif
(d) Offered memory bandwidth (references/sec).
d87111c01013bcda00bb8640fdff6754.gif
(e) Achieved memory bandwidth (references/sec) according to above model.
d87111c01013bcda00bb8640fdff6754.gif
(f) Achieved memory bandwidth using Strecker's model.
6. The processor defined in problem 4 now uses an integrated memory system of eight modules.
d87111c01013bcda00bb8640fdff6754.gif
Compare performance (B, Bw, Qt, and Tw) projections based on:
d87111c01013bcda00bb8640fdff6754.gif
(a) Hellerman's model.
d87111c01013bcda00bb8640fdff6754.gif
(b) Strecker's model.
d87111c01013bcda00bb8640fdff6754.gif
(c) Open-queue model.
d87111c01013bcda00bb8640fdff6754.gif
(d) Simple closed binomial model.
d87111c01013bcda00bb8640fdff6754.gif
Compute achieved bandwidth, buffer size, and Tw for each; state all assumptions.
7. Evaluate the processor defined in problem 4 with single integrated memory and m = 8, using the d-binomial model. Find the achieved processor performance.

 
< previous page page_421 next page >