Memory Hierarchy in Computer Systems - Prof. Leona Mcmillan, Study notes of Computer Architecture and Organization

Download Memory Hierarchy in Computer Systems - Prof. Leona Mcmillan and more Study notes Computer Architecture and Organization in PDF only on Docsity! L2 1 – M em o ry H ie ra rc hy 1 C o m p 4 11 – F a ll 2 0 0 7 11 /1 4 /0 7 M e m or y H ie ra rc h y W hy a re y o u d re s s ed lik e th a t? H a llo w ee n w a s w ee ks a go ! It m a ke s m e lo o k fa s te r, d o n’ t yo u th in k? •M em o ry F la vo rs •P ri n ci p le o f Lo ca lit y •P ro gr a m T ra ce s •M em o ry H ie ra rc h ie s •A s s o ci a ti vi ty (S tu d y C h a p te r 7) L2 1 – M em o ry H ie ra rc hy 2 C o m p 4 11 – F a ll 2 0 0 7 11 /1 4 /0 7 W h a t D o W e W a nt i n a M e m or y ? P C IN S T M A D D R M D A T A m in iM IP S M E M O R Y ¢ 10 m s 10 0 ’s G by te s H a rd d is k* W a nt ? $ 5 n s 10 0 ’s M by te s D R A M $$ $ 0 .2 n s 10 0 ’s K by te s S R A M $$ $$ 10 p s 10 0 0 ’s o f bi ts R eg is te r C o s t La te n cy C a p a ci ty * no n- vo la ti le A D D R D O U T A D D R D A T A R /W W r 2 G by te 0 .2 n s c he a p L2 1 – M em o ry H ie ra rc hy 5 C o m p 4 11 – F a ll 2 0 0 7 11 /1 4 /0 7 Q ua nt it y v s Q ua li ty … Y o u r m em o ry s ys te m c a n b e • B IG a n d S LO W ... o r • S M A LL a n d F A S T. 10 -8 10 -3 1 10 0 .110 10 0 0 10 0 1 10 -6 D VD B ur ne r (0 .0 2 $/ G B , 1 2 0 m s ) D IS K ( 0 .0 5 $/ G B , 1 0 m S ) D R A M ( 2 5 $/ G B , 5 n s ) S R A M ( 10 0 0 $/ G B , 0 .2 n s ) A cc es s Ti m e .0 1 $/ G B W e’ ve e xp lo re d a r a ng e of d ev ic e- d es ig n tr a d e- o ff s . Is t h er e a n A R C H IT E C TU R A L s o lu ti o n to t h is D E LI M A ? 1 L2 1 – M em o ry H ie ra rc hy 6 C o m p 4 11 – F a ll 2 0 0 7 11 /1 4 /0 7 M a na gi ng M e m or y v ia P ro gr a m m in g • In r ea lit y, s ys te m s a re b u ilt w it h a m ix tu re o f a ll th es e va ri o u s m em o ry t yp es • H o w d o w e m a ke t h e m o s t ef fe ct iv e u s e of e a ch m em o ry ? • W e co u ld p u s h a ll o f th es e is s u es o ff t o p ro gr a m m er s • K ee p m o s t fr eq ue nt ly u s ed v a ri a bl es a nd s ta ck in S R A M • K ee p la rg e d a ta s tr uc tu re s ( a rr a ys , l is ts , e tc ) in D R A M • K ee p b ig ge r d a ta s tr uc tu re s o n d is k (d a ta ba s es ) on D IS K • It is h a rd er t h a n y o u t h in k… d at a u s a ge e vo lv es o ve r a p ro gr a m ’s e xe cu ti o n C P U S R A M M A IN M E M L2 1 – M em o ry H ie ra rc hy 7 C o m p 4 11 – F a ll 2 0 0 7 11 /1 4 /0 7 B e st o f B ot h W or ld s W h a t w e R E A LL Y w a nt : A B IG , F A S T m em o ry ! (K ee p e ve ry th in g w it h in in s ta nt a cc es s ) W e’ d li ke t o h a ve a m em o ry s ys te m t h a t • P E R F O R M S li ke 2 G B yt es o f S R A M ; b ut • C O S TS li ke 5 12 M B yt es o f s lo w m em o ry . S U R P R IS E : W e ca n ( n ea rl y) g et o u r w is h ! K E Y : U s e a h ie ra rc h y o f m em o ry t ec h n o lo gi es : C P U S R A M M A IN M E M L2 1 – M em o ry H ie ra rc hy 10 C o m p 4 11 – F a ll 2 0 0 7 11 /1 4 /0 7 W or ki ng S e t ti m e a d d re s s d a ta s ta c k p ro g ra m ∆ t |S | ∆ t S is t he s et o f lo ca ti o ns a cc es s ed d ur in g ∆ ∆∆∆ t. W o rk in g s et : a s et S w hi ch c ha ng es s lo w ly w .r .t . a cc es s t im e. W o rk in g s et s iz e, |S | L2 1 – M em o ry H ie ra rc hy 11 C o m p 4 11 – F a ll 2 0 0 7 11 /1 4 /0 7 E x pl oi ti ng t h e M e m or y H ie ra rc h y A p p ro a ch 1 ( C ra y, o th er s ): E xp o s e H ie ra rc h y • R eg is te rs , M a in M em o ry , D is k ea ch a va ila bl e a s s to ra ge a lt er na ti ve s ; • Te ll p ro gr a m m er s : “ U s e th em c le ve rl y” A p p ro a ch 2 : H id e H ie ra rc hy • P ro g ra m m in g m o d el : S IN G LE k in d o f m em o ry , s in gl e a d d re s s s p a ce . • M a ch in e A U TO M A TI C A LL Y a s s ig ns lo ca ti o ns t o f a s t o r s lo w m em o ry , d ep en d in g o n u s a ge p a tt er ns . C P U S R A M M A IN M E M C P U S m a ll S ta ti c D yn a m ic R A M H A R D D IS K “M A IN M E M O R Y ” L2 1 – M em o ry H ie ra rc hy 12 C o m p 4 11 – F a ll 2 0 0 7 11 /1 4 /0 7 W h y W e C a re C P U S m a ll S ta ti c D yn a m ic R A M H A R D D IS K “M A IN M E M O R Y ” TR IC K # 1: H ow t o m a ke s lo w M A IN M E M O R Y a p p ea r fa s te r th a n it is . C P U p er fo rm a n ce is d o m in a te d b y m em o ry p er fo rm a n ce . M o re s ig ni fi ca nt t ha n: IS A , c ir cu it o pt im iz a ti on , p ip el in in g, s up er -s ca la r, e tc TR IC K # 2 : H ow t o m a ke a s m a ll M A IN M E M O R Y a p p ea r bi gg er t ha n it is . “V IR TU A L M E M O R Y ” “S W A P S P A C E ” Te ch ni q ue : VI R TU A L M E M O R Y – Le ct ur e a ft er t ha t “C A C H E ” Te ch ni q ue : C A C H E IN G – N ex t 2 L ec tu re s L2 1 – M em o ry H ie ra rc hy 15 C o m p 4 11 – F a ll 2 0 0 7 11 /1 4 /0 7 T h e C a ch e P ri nc ip le F in d “ H a rt , L ee ” 5 -M in ut e A cc es s T im e: 5 -S ec on d A cc es s T im e: A LG O R TH IM : L o o k on y o ur d es k fo r th e re q ue s te d in fo rm a ti on f ir s t, if it s n o t th er e ch ec k s ec on d a ry s to ra ge L2 1 – M em o ry H ie ra rc hy 16 C o m p 4 11 – F a ll 2 0 0 7 11 /1 4 /0 7 B a si c C a ch e A lg or it h m O N R E F E R E N C E T O M em [X ]: L o o k fo r X a m o ng c a ch e ta gs ... H IT : X = T A G (i ) , f o r s o m e ca ch e lin e i R E A D : re tu rn D A TA (i ) W R IT E : ch a ng e D A TA (i ); S ta rt W ri te t o M em (X ) M IS S : X n o t fo u n d in T A G o f a n y ca ch e lin e R E P LA C E M E N T S E LE C TI O N : S el ec t s o m e LI N E k t o h o ld M em [X ] (A llo ca ti o n ) R E A D : R ea d M em [X ] S et T A G (k )= X, D A TA (K )= M em [X ] W R IT E : S ta rt W ri te t o M em (X ) S et T A G (k )= X, D A TA (K )= n ew M em [X ] M A IN M E M O R Y C P U (1 − α ) Ta g D a ta A B M em [A ] M em [B ] “X ” h er e is a m em o ry a d d re s s . L2 1 – M em o ry H ie ra rc hy 17 C o m p 4 11 – F a ll 2 0 0 7 11 /1 4 /0 7 C a ch e S it s b et w ee n C P U a nd m a in m em o ry Ve ry f a s t m em o ry t h a t s to re s T A G s a n d D A TA TA G is t h e m em o ry a d d re s s ( o r p a rt o f it ) D A TA is a c o p y of m em o ry a t th e a d d re s s g iv en b y TA G 11 10 0 8 9 7 10 3 2 1 10 4 0 17 10 0 0 4 10 4 4 1 10 4 0 2 5 10 3 6 9 7 10 3 2 9 9 10 2 8 4 4 10 2 4 3 8 10 2 0 2 9 10 16 5 10 12 11 10 0 8 2 3 10 0 4 17 10 0 0 M e m o ry T a g D a ta C a c h e L in e 0 L in e 1 L in e 2 L in e 3 L2 1 – M em o ry H ie ra rc hy 2 0 C o m p 4 11 – F a ll 2 0 0 7 11 /1 4 /0 7 H ow d o w e S e a rc h t h e C a ch e T A G s? N op e, “ S m it h” N op e, “ J on es ” N op e, “ Le Vi le ” H E R E IT IS ! F in d “ H a rt , L ee ” A s s o ci a ti vi ty : Th e d eg re e o f p a ra lle lis m u s ed in th e lo o ku p o f Ta gs L2 1 – M em o ry H ie ra rc hy 2 1 C o m p 4 11 – F a ll 2 0 0 7 11 /1 4 /0 7 F ul ly - A ss oc ia ti ve C a ch e TA G D a ta = ? TA G D a ta = ? TA G D a ta = ? In co m in g A d d re s s H IT D a ta O ut Th e ex tr em e in a s s o ci a ti ve ly : A ll TA G S a re s ea rc he d in p a ra lle l D a ta it em s f ro m * a ny * a d d re s s c a n be lo ca te d in *a ny * ca ch e lin e L2 1 – M em o ry H ie ra rc hy 2 2 C o m p 4 11 – F a ll 2 0 0 7 11 /1 4 /0 7 D ir e ct - M a pp e d C a ch e (n on - a ss oc ia ti ve ) N O P a ra lle lis m : Lo o k in J U S T O N E p la ce , d et er m in ed b y p a ra m et er s o f in co m in g re q u es t (a d d re s s b it s ) ... c a n u s e o rd in a ry R A M a s ta bl e A F in d “ H a rt , L ee ” Y Z B H L2 1 – M em o ry H ie ra rc hy 2 5 C o m p 4 11 – F a ll 2 0 0 7 11 /1 4 /0 7 D ir e ct M a pp e d C a ch e Lo w -c o s t ex tr em e: S in gl e co m p a ra to r U s e o rd in a ry ( fa s t) s ta ti c R A M f o r ca ch e ta gs & d a ta : In co m in g A d d re s s K T = ? H IT D a ta O ut D IS A D VA N TA G E : C O LL IS IO N S Q U E S TI O N : W hy n o t us e H IG H -o rd er bi ts a s t he C a ch e In d ex ? K -b it C a ch e In d ex D -b it d a ta w o rd T U p p er -a d d re s s b it s Ta g D a ta K x ( T + D )- bi t s ta ti c R A M L2 1 – M em o ry H ie ra rc hy 2 6 C o m p 4 11 – F a ll 2 0 0 7 11 /1 4 /0 7 A P ro b le m w it h C ol li si on s F in d “ H ee l, A rt ” F in d “ H er e, A l T .” F in d “ H a rt , L ee ” N o p e, I’ ve g o t “H ee l” u n d er “ H ” P R O B LE M : C o nt en ti o n a m o n g H ’s ... . - C A N ’T c a ch e bo th “H a rt ” & “ H ee l” ... S u p p o s e H ’s t en d to c o m e a t o n ce ? = = > B E TT E R ID E A : F ile b y LA S T le tt er ! Y Z B H L2 1 – M em o ry H ie ra rc hy 2 7 C o m p 4 11 – F a ll 2 0 0 7 11 /1 4 /0 7 C a ch e Q ue st io ns = C a sh Q ue st io ns W h a t lie s b et w ee n F u lly A s s o ci a te a n d D ir ec t- M a p p ed ? W h en I p ut s o m et hi n g n ew in to t he c a ch e, w h a t d a ta g et s th ro w n o ut ? H o w m a ny p ro ce s s o r w o rd s s h o u ld t h er e be p er t a g? W h en I w ri te t o c a ch e, s h o u ld I a ls o w ri te t o m em o ry ? W h a t d o I d o w he n a w ri te m is s es c a ch e, s h o u ld s p a ce in ca ch e be a llo ca te d f o r th e w ri tt en a d d re s s . W h a t if I h a ve IN P U T/ O U TP U T d ev ic es lo ca te d a t ce rt a in m em o ry a d d re s s es , d o w e ca ch e th em ?