You are on page 1of 13
Exercises Les rachines wil pe anizations, Assume i a" in men ft for Ul he juiTes 200 ns (Jor th f oes not all " 0 Bindi : A h f he program : E Hi fhe program 1212) f , Th ht ram st n ierarchy. B nC for UNIX at uses a standard UNE q ne. T below was taken fr He oM 471024) /* largest ca 7 r to get a larger tine sa T LCTCK 60 er clock ticks per 4 int x{c 1 ping to stride through +, times (ru uN me in clock ti urn (do 7 ut ime) / nt steps, tstep double se ng i 'e i tri d r E*stri 0 jer samy 0 03 ind index=indextstrsd ex] = xfindex] + ne ec = see + (get_sec: = sect d tin 0 ed to mat ter ae single % = sec - (ge nd = 9 h tep 7 e * "i d 74 readtwrite:14.0f ns\n' ze ” zeof( ) ect le H mit ) program that program ad rick physical addre Figure 5.59 Results of running program in fon a Sun Blade 100¢ 1 of a split c the [ h Th ¥ all rit r ns at “ h a ne that has no hit penalty, an mais the same in abs fe simulator to see if Sm sults hol for meme ince traces that you ha IW they donot hold, why 5 ’ 4 that the best memory hierarch r 7 possible to prevent som tions fro lst plain why Mef ult could be t | 5.2, $.S> The four m . tions (S 2) form " f x . he 7 or not fit into this provide the hit time of u direct-mapp MIPS RIK proc r > ach erent posl: re 1 of the chip pac The RIOK has nel ip Lic mn-chip L2 tag comparison cirouitry, and an on-chip L predic L2 tag information is brough in 7 1 se ble contains SK I Neessor p must be te soci ma © sdiction m vins needed the RIOK and data, and impact on p mance compared (0 a pat full complement of pins to inter the L2 cach b. LISI is the performan just using th nal pins hut ot inc i y " ne RIOK 0 ly use lic What are reasonable design ch fe state update(s) 10 i J data is ith xd way. the desire nthe ne i Jata isnot in the secondary eacla . 15 a ache hay € seks, how 1 pe mnt rics are needed?” How would the RLOK support 5 44 MB L2 cache with 12 pow is the usel RINK way prediction table analogous to that of a branch Simulating a os niss rates that a ui mpl a “ nl produces a stream nces that more accuratel ” puter workload, SimpleSe precomp! SP Exerci 523 y u r hs blocks, what te an he measured 7 u Is : ; h Jn the range of py miss rate? lise 2. havi nmplete provra billions and bili i i resou Under certain assump sit nulation uh " hile main m f Th if mit 1 5.16 r me and effort, tre o i vot shorter o 1 v0 " impossible 10 a miss s h 24 The mis ata report ‘i 7 1 ule a trace 8 vi ation? F : nun factor ala Juced? In f ated (esse " ction int si neoded n decision nne design q made placemen i ia point during trace si the ed nts OF sek ay a i nave the same block size and num it the k " un mm “least reph replaceat fi h fe precisely the bh nl be bel at ex nd their order repres " 1 re 7 : ull asc Jock. Thus, the ble i be i h ing atthe top. all " RL i ing replacement ied ti ibe lacement x « ibe an plish this econ 7 i Hata structure lata re ma

You might also like