Sie sind auf Seite 1von 37

'o(ert Cattera))& IBM rf atter*$s+i(m+ om

Memory for MIPS: Leveraging Big Memory on System z to Enhan e !B2 CP" Effi ien y
Mi higan !B2 "sers #ro$p May 1%& 2012

Information Management

2013 IBM Corporation

Information Management

,gen-a
.he $rrent )an-s ape #etting yo$r ($ffer poo) ho$se in or-er Being (o)- / ($t not re 0)ess / in as0ing for more rea) storage for a !B2 s$(system 1e2 opport$nities for e3p)oiting 'ELE,SE4!E,LL5C,.E6 5ther 2ays to tra-e memory for MIPS

2013 IBM Corporation

Information Management

.he $rrent )an-s ape

2013 IBM Corporation

Information Management

System z server memory: getting BI#


Pro-$ tion z85S LP,'s 2ith rea) storage reso$r es of 20 #B& 70 #B& or more are in reasing)y ommon
/ Seeing some pro-$ tion z85S LP,'s 2ith 1009 #B of rea) storage

:hat;s -riving the tren- to2ar-s )arger z85S LP,' memory sizes<
/ .he -esire for =(a)an e-> onfig$rations
? Pro essing apa ity 4(ase- on n$m(er an- spee- of engines in z85S images6 0eeps gro2ing ? zEC12 server: $p to 101 engines& ea h pa 0ing a(o$t 1000 MIPS of apa ity ? 1ot $n$s$a) to see 10@20 #B of rea) storage per engine

/ 'e-$ e- ost of System z memory

AA

2013 IBM Corporation

Information Management

1e2 2ays to )everage Big Memory on z


'e ent !B2 for z85S -eve)opments provi-e more opport$nities to $se System z memory a-vantageo$s)y , fe2 e3amp)es 4 overe- in more -etai) )ater in presentation6:
/ Page@fi3e- ($ffer poo)s 4an-& 2ith !B2 10& $se of 1 MB page frames for s$ h poo)s6 / !B2@a2are =pinning> of o(Be ts in ($ffer poo)s 4!B2 106 / .hrea-@re)ate- virt$a) storage a)most entire)y a(ove 2 #B =(ar> 2ith !B2 10 42hen pa 0ages (o$n- or re(o$n- in !B2 10 environment6
? More on $rrent threa-s per s$(system ? More =hea- room> for $se of 'ELE,SE4!E,LL5C,.E6 (in- option

/ Cigh@performan e !B,.s 4!B2 106 / tie- to 'ELE,SE4!E,LL5C,.E6

2013 IBM Corporation

Information Management

.here are a )ot of =s)eeping giga(ytes> o$t there


,t many !B2 for z85S sites& L5.S of spare memory apa ity !o yo$ 0no2 yo$r z85S system;s -eman- paging rate<
/ ,vai)a()e via a z85S monitor& this is the rate at 2hi h pages that have (een sent (y z85S to a$3i)iary storage are page- (a 0 into system memory on@-eman/ In my e3perien e& it;s often )ess than 1 per se on-& even -$ring ($sy perio-s
? If the -eman- paging rate is in the )o2 sing)e -igits 4or )ess6 per se on-& z85S LP,' memory is not stresse- / look for opportunities to use more of it as a means of boosting the CPU efficiency of your DB2 workload

Zzzzz

2013 IBM Corporation

Information Management

I a)) this memory for MIPS& an- that;s 2hat this presentation is a)) a(o$t

2013 IBM Corporation

Information Management

#etting yo$r ($ffer poo) ho$se in or-er

2013 IBM Corporation

Information Management

:hat I meanG
Before en)arging a !B2 s$(system;s ($ffer poo) onfig$ration& ma0e s$re that yo$;re getting the most o$t of the onfig$ration as $rrent)y size-

2013 IBM Corporation

Information Management

Hirst& $stomize settings for 2or0 fi)e ($ffer poo)s


'eferring here to the ($ffer poo)s -e-i ate- to the 7I@page an32I@page 2or0 fi)e ta()e spa es
/ Jo$ !5 assign 2or0 fi)e ta()e spa es to their o2n ($ffer poo)s& -on;t yo$<

:or0 fi)e ta()e spa es are -ifferent from others in a o$p)e of 2ays that have imp)i ations for re ommen-e- ($ffer poo) parameter settings
/ Hor one thing& a)most a)) rea-s are of the prefet h variety
? :hy that matters: the -efa$)t va)$e of the KPSEL. ($ffer poo) parameter 4the per entage of a poo);s pages that an (e o $pie- (y pages rea- into memory via prefet h6 is F0 ? Stay 2ith that& an- yo$;)) (e 2asting 20M of the ($ffers in a poo) -e-i ate- to 2or0 fi)e ta()e spa es ? In reasing KPSEL. to A%@AAM for a 2or0 fi)e@-e-i ate- poo) sho$)- res$)t in -e rease- rea- I85 a tivity& an- fe2er I85s means )ess CP" ons$mption
10 2013 IBM Corporation

Information Management

:or0 fi)e@-e-i ate- poo)s / the other -ifferen e


!ifferent motivation for e3terna)izing hange- pages to -is0
/ Hor other ($ffer poo)s& getting hange- pages e3terna)ize- to -is0 in a time)y manner is important for !B2 restart performan e
? Hor2ar- )og re overy phase of restart: -ata on -is0 $p-ate- to ref)e t hanges ommitte-& ($t not yet e3terna)ize-& at the time of !B2 s$(system fai)$re ? .he more hange-@($t@not@yet@e3terna)ize- pages there are in ($ffer poo)s 4other than 2or0 fi)e poo)s6 at time of !B2 fai)$re& the )onger restart 2i)) ta0e ? .hat (eing the ase& yo$ 2ant fair)y )o2 va)$es for -eferre- 2rite thresho)-s 4!:L.& K!:L.6 for these poo)s 4-efa$)ts of 30 an- % are $s$a))y goo-6

/ :or0 fi)es are )i0e s rat h pa-s for N$ery pro essing / they are not re overe- 2hen !B2 is restarte- 4N$eries 2o$)- (e res$(mitte-6
? .herefore& on)y nee- a )eve) of hange-@page e3terna)ization a tivity that is s$ffi ient to prevent thrashing 4i+e+& avoi- shortage of stea)a()e ($ffers6 ? He2er page 2rites O )ess CP" ons$mption ? I;ve seen E0870 $se- for !:L.8K!:L.& an- a)so F08%0 ? But dont take this too far
11 2013 IBM Corporation

Information Management

, 2arning a(o$t too@high !:L.8K!:L. va)$es


,t one site& I sa2 va)$es of A0 for (oth !:L. an- K!:L. for a 2or0 fi)e@-e-i ate- ($ffer poo) .he -ata manager thresho)- 4a0a !M.C6 is rea he- 2hen A%M of a poo);s ($ffers are non@stea)a()e 4either $rrent)y in $se or hange-@($t@not@yet@e3terna)ize-6
/ :hen !M.C is hit for a poo)& there 2i)) (e a #E.P,#E for ea h ro2 retrieve- from a page a he- in that poo)
? So& if 20 ro2s are retrieve- from a page& there 2i)) (e 20 #E.P,#Es vs+ 1 ? 'es$)t: signifi ant in rease in !B2 CP" ons$mption

/ :ith !:L. an- K!:L. (oth at A0& at this site !M.C 2as hit many times per ho$r for the ($ffer poo) in N$estion& $n(e0no2nst to !B2 staff / Che 0 !B2 monitor -isp)ay or statisti s report& or o$tp$t of !B2 omman- /!ISPL,J B"HHE'P55L4BPn6 !E.,IL to see if !M.C hit
? Jo$ a)2ays 2ant that va)$e to (e zero
12 2013 IBM Corporation

Information Management

More goo- ($ffer ste2ar-ship: stea) smart


By -efa$)t& !B2 $ti)izes a )east@re ent)y@$se- 4L'"6 a)gorithm for ($ffer stea)ing 4i+e+& to se)e t o $pie- ($ffers that 2i)) (e over2ritten to a ommo-ate ne2 pages (ro$ght in from -is06
/ .his is the right a)gorithm to $se in most ases

In a pre@!B2 10 system& if yo$ 2ant to $se a ($ffer poo) to =pin> -ata(ase o(Be ts in memory 4i+e+& to have those o(Be ts a hein memory in their entirety6& $se the HIH5 ($ffer stea) a)gorithm
/ :hy< Be a$se HIH5 4first in& first o$t6 is simp)er than L'" an- so $ses )ess CP" / If o(Be ts are a he- in their entirety 4i+e+& if tota) n$m(er of pages for o(Be ts assigne- to poo) is )ess than n$m(er of ($ffers in poo)6& there 2on;t (e any ($ffer stea)ing& so 2hy spen- CP" tra 0ing ($ffer $sage< / DB2 10 provided a new and better way to do this
13 2013 IBM Corporation

Information Management

Smarter pinning 2ith !B2 10


1e2 ($ffer stea) a)gorithm for poo)s: P#S.E,L4151E6 :hen P#S.E,L4151E6 is spe ifie- for a poo):
/ :hen an o(Be t 4ta()e spa e or in-e36 assigne- to the poo) is first a esse-& reN$esting app)i ation pro ess 2i)) get 2hat it nee-s an- !B2 2i)) asyn hrono$s)y rea- a)) the rest of the o(Be t;s pages into the poo) / Hor s$(seN$ent SLL statements targeting o(Be t that is no2 pinne- in the poo)& !B2 2i)) ass$me that no I85s 2i)) (e nee-e-& an- 2i)) optimize the SLL statements a or-ing)y / If o(Be ts assigne- to a P#S.E,L4151E6 poo) have more pages than the poo) has ($ffers& !B2 2i)) a$tomati a))y s2it h to HIH5 ($ffer stea) a)gorithm to a ommo-ate ne2 page rea-s from -is0 2hen poo) is f$))

17

2013 IBM Corporation

Information Management

Be smart a(o$t page@fi3ing ($ffers& too


P#HIP4JES6 spe ifi ation for a ($ffer poo) saves MIPS (y ma0ing -ata(ase rea- an- 2rite I85s )ess ost)y
/ P#HIP4JES6 ($ffers stay fi3e- in memory& so no nee- for !B2 to as0 z85S to fi3 4an- s$(seN$ent)y re)ease6 a rea) storage page frame ho)-ing a ($ffer 2hen -ata is rea- into& or 2ritten from& that ($ffer
? Hor a prefet h rea- of 32 pages& that;s 32 page@fi38page@re)ease operations avoi-e-

Sin e CP" effi ien y (enefit 4prior to !B2 106 is re)ate- to I85 a tivity& P#HIP4JES6 2on;t have m$ h of an impa t for a poo) that sho2s on)y a fe2 rea- I85s per se on/ 'e ommen-ation: $se P#HIP4JES6 for poo)s that have a )ot of rea- I85 a tivity 4e+g+& 100s or 1000s per se on-6& an- P#HIP4151E6 for )o2@I85 poo)s 4to give z85S some stea)a()e reso$r es if the nee- arises6
? I;)) over monitoring of rea- I85 rates for poo)s momentari)y
1% 2013 IBM Corporation

Information Management

!B2 10: ne2 P#HIP4JES6 (enefit


:ith z10& z1AD& an- zEC12 servers 4z85S 1+10 or )ater6& part of memory reso$r e an (e manage- in 1 MB page frames
/ LH,'E, parm in IE,SJS33 mem(er of P,'MLIB

!B2 10 2i)) $se 1 MB page frames to (a 0 a P#HIP4JES6 ($ffer poo) 4if not eno$gh 1 MB page frames to f$))y (a 0 poo)& !B2 10 2i)) $se 7 IB frames in a--ition to 1 MB frames6
/ Larger page frame means more CP"@effi ient trans)ation of virt$a) storage a--resses to rea) storage a--resses / .his CP" (enefit is on top of the previo$s)y mentione- MIPS savings res$)ting from )ess@ ost)y I85s into an- o$t of page@fi3e- ($ffer poo)s

!B2 KF an- KA an;t $se 1 MB page frames for ($ffer poo)s


/ So& 2hen going to !B2 10 -on;t onfig$re an over)y )arge part of server memory to (e manage- 2ith 1 MB page frames $nti) yo$;re pretty s$re yo$ 2on;t have to fa)) (a 0
1D 2013 IBM Corporation

Information Management

'e-$ ing I85s (y re-istri($ting ($ffers


'eferring here to re-$ ing size of )o2@I85 poo)s an- in reasing size of high@I85 poo)s (y a )i0e amo$nt
/ #oa): a(o$t same rea- I85 rate for re-$ e-@size poo)s 4-on;t ta0e too many ($ffers from these poo)s6& an- )ess rea- I85 a tivity for en)argepoo)s / a)) 2ith no net in rease in overa)) size of ($ffer poo) onfig$ration
? ,gain& fe2er I85s O re-$ e- CP" ons$mption

Step 1: -etermine rea- I85 rate for ea h ($ffer poo)& $sing information o(taine- from:
/ !B2 monitor 4statisti s report& or on)ine -isp)ay of ($ffer poo) a tivity6 / @or@ / 5$tp$t of !B2 omman- @!ISPL,J B"HHE'P55L4,C.IKE6 !E.,IL
? Iss$e omman- on e& then 2ait an ho$r an- iss$e it again ? 5$tp$t of se on- iss$an e of omman- 2i)) apt$re 1 ho$r of a tivity / -ivi-e a tivity o$nters (y 3D00 to get per@se on- fig$res
1E 2013 IBM Corporation

Information Management

Ca) $)ating a poo);s rea- I85 rate


:hat yo$ 2ant is per@se on- -ata .ota) rea- I85 rate is s$m of:
/ a)) syn hrono$s rea-s 4ran-om an- seN$entia)6 per se on/ an/ a)) prefet h rea-s 4seN$entia) prefet h 9 )ist prefet h 9 -ynami prefet h6 per se on-

E3amp)e: BP1 an- BP2 (oth have 70&000 ($ffers& an- tota) reaI85 rate is 208se on- for BP1 an- 20008se on- for BP2
/ In that ase& I;- onsi-er -e reasing size of BP1 (y 20&000 ($ffers anin reasing size of BP2 (y 20&000 ($ffers

1F

2013 IBM Corporation

Information Management

Being (o)- / ($t not re 0)ess / in as0ing for more rea) storage for a !B2 s$(system

1A

2013 IBM Corporation

Information Management

:o$)- (igger (e (etter<


If yo$r ($ffer poo) ho$se is in or-er& sho$)- yo$ ma0e the ho$se (igger 4i+e+& in rease the size of the overa)) ($ffer poo) onfig$ration6<
/ !epen-s / 2hat;s the $rrent tota) rea- I85 rate for ea h of yo$r ($ffer poo)s 4see s)i-e 1F6<
? If rea- I85 rate for ea h ($ffer poo) is )ess than 100 per se on-& ma0ing the poo)s )arger is not )i0e)y to have m$ h of a CP"@saving impa t ? If rea- I85 rate for one or more poo)s is in the 100s or 1000s per se on-& ma0ing that poo) 4or poo)s6 )arger o$)- yie)- signifi ant CP" effi ien y (enefits 4(e a$se fe2er I85s O )ess CP" ons$mption6

20

2013 IBM Corporation

Information Management

If yo$;re going to ma0e a ($ffer poo) (iggerG


Ga-- eno$gh ($ffers to ma0e a -ifferen e
/ ,--ing another 1000 ($ffers to a poo) that a)rea-y has F0&000 ($ffers is not )i0e)y to move the nee-)e very m$ h
? In rease the size of that F0&000@($ffer poo) (y 20&000 or 70&000 ($ffers& an- I;- say& =1o2 yo$;re ta)0in> ? If a poo) is N$ite sma)) / say& 10&000 7I ($ffers / an- has a high rea- I85 rate& I might 2ant to in rease its size (y a fa tor of t2o 4or more6

Some organizations are starting to =get it&> in terms of ta0ing a-vantage of Big Memory to re-$ e !B2 CP" ons$mption via )arger ($ffer poo) onfig$rations
/ .he )argest ($ffer poo) onfig$ration I;ve seen on one !B2 for z85S s$(system is 72 #B

21

2013 IBM Corporation

Information Management

#et on the same page as yo$r z85S sysprog


,gain& the meas$re of =memory stress> that I )i0e to $se is the -eman- paging rate
/ ,s yo$ imp)ement memory@for@MIPS hanges& 0eep an eye on the z85S LP,';s -eman- paging rate& an- -on;t )et this get o$t of han/ If the -eman- paging rate is in the sing)e -igits 4or )ess6 per se on-& it;s not o$t of han-

:hatever yo$r LP,';s -eman- paging rate& I;- (e aref$) a(o$t $sing more than %0M of an LP,';s memory for !B2 ($ffer poo)s
/ I;ve seen :LM size a ($ffer poo) onfig$ration at 30@70M of LP,' memory 4,L.E' B"HHE'P55L4BPn6 ,".5SIQE4JES66

22

2013 IBM Corporation

Information Management

Ieep monitoring rea- I85 rate for ($ffer poo)s


I genera))y ta0e a triage approa h: fo $s efforts on the poo)s 2ith the highest rea- I85 rates
/ Cighest rate I;ve seen is A000 rea- I85s per se on- for one ($ffer poo)

My aim: if possi()e& get rea- I85 rate to )ess than 1000 per se on- for ea h ($ffer poo)
/ If that o(Be tive is a omp)ishe- 4or if I;m seeing -iminishing ret$rns 2ith respe t to en)arging a high@I85 ($ffer poo)6& I t$rn my fo $s to poo)s 2ith rea- I85 rates (et2een 100 an- 1000 per se on? 1i e to get these (e)o2 100 per se on-& if possi()e

Ieep in min-: it;s not B$st a(o$t ma0ing e3isting poo)s )arger
/ ,t some point& yo$ may 2ant to reate a ne2 BPy& an- reassign o(Be ts to that poo) from BP3
? Can (e parti $)ar)y effe tive for separating =history> vs+ = $rrent> ta()es& a ess patterns for 2hi h ten- to (e -ifferent
23 2013 IBM Corporation

Information Management

If -ata sharing& -on;t forget gro$p ($ffer poo)s


Sometimes peop)e 2i)) ma0e BP3 )arger a ross mem(ers of a -ata sharing gro$p& an- 2i)) forget to en)arge #BP3 a or-ing)y
/ If aggregate size of )o a) BPs gets too )arge re)ative to the size of the orrespon-ing #BP& yo$ o$)- en- $p 2ith a )ot of -ire tory entry re )aims an- that;s not goo- for performan e
? Can he 0 on -ire tory entry re )aim a tivity $sing o$tp$t of !B2 omman@!ISPL,J #'5"PB"HHE'P55L4#BPn6 #!E.,IL

Hor a 7I #BP 2ith the -efa$)t %:1 ratio of -ire tory entries to -ata pages& -ire tory entry re )aims )i0e)y to (e 0 if size of #BP is 70M of om(ine- size of asso iate- )o a) BPs
/ E3amp)e: 3@2ay gro$p& BP1 at 70I ($ffers 41D0 MB6 on ea h mem(er
? #oo- #BP1 size is 70M P 43 P 1D0 MB6 O 1A2 MB
27 2013 IBM Corporation

Information Management

1e2 opport$nities for e3p)oiting 'ELE,SE4!E,LL5C,.E6

2%

2013 IBM Corporation

Information Management

:hat;s goo- a(o$t 'ELE,SE4!E,LL5C,.E6<


Pa 0age (in- option& a$ses !B2 to retain ertain reso$r es a))o ate- to a threa- 4ta()e spa e@)eve) )o 0s& E!M poo) e)ements6 $nti) threa- -ea))o ation& vs+ retaining $nti) ommit 'ELE,SE4!E,LL5C,.E6 an save MIPS vers$s 'ELE,SE4C5MMI.6 2hen:
/ .hrea- in N$estion persists thro$gh ommits 4e3amp)es are (at h threa-s an- CICS prote te- entry threa-s6 / @an-@ / .hrea- is $se- repeate-)y for the e3e $tion of the same pa 0age
? In that ase& 'ELE,SE4!E,LL5C,.E6 avoi-s CP" ost of repeatere)ease an- re@a N$isition of same ta()e spa e )o 0s an- E!M poo) e)ements at ea h ommit

2D

2013 IBM Corporation

Information Management

:hat;s the memory@for@MIPS ang)e<


'ELE,SE4!E,LL5C,.E6 in reases memory $ti)ization (e a$se it in reases amo$nt of virt$a) storage $se- (y threa-s Before !B2 10& a signifi ant portion of threa-@re)ate- virt$a) storage 2as a N$ire- (e)o2 the 2 #B =(ar> in !BM1
/ .a)0ing a(o$t the part of the E!M poo) $se- for C. 4i+e+& $rsor ta()e& if !B'M (o$n- -ire t)y into p)an6 an- P. 4pa 0age ta()e6
? Copies of pa 0age an- $rsor ta()e se tions a))o ate- to threa-s

/ !B2 monitor might refer to this part of E!M poo) as =Pages in '!S Poo)> 4a(ove an- (e)o2 the (ar6& in E!M poo) se tion of statisti s report / "se of 'ELE,SE4!E,LL5C,.E6 ma0es more of this spa e non@ stea)a()e& 2hi h o$)- potentia))y )ea- to program fai)$re -$e to )a 0 of spa e in E!M poo)
? #enera) re ommen-ation: per entage of free pages in this part of the E!M sho$)- (e at )east 10M
2E 2013 IBM Corporation

Information Management

!B2 10 threa- storage


E%@A0M )ess $sage of (e)o2@ the@(ar !BM1 virt$a) storage 2ith !B2 10 vs+ !B2 A
/ Primari)y -$e to movement of a)most a)) threa-@re)ate- storage a(ove the 2 #B (ar

#)o(a) !SC !B! Lo a) !SC

C.8P. SIC.

.hrea- 8 Sta 0 SIP. .hrea- 8 Sta 08 2or0ing E%@A0M )ess $sage !BM1 (e)o2 (ar after 'EBI1!

'es$)t: m$ h more virt$a) storage =hea- room> for $se of 'ELE,SE4!E,LL5C,.E6 ,)so& QP,'M no )onger )imits spa e avai)a()e for pa 0age ta()e / no2 pretty m$ h )imite(y avai)a(i)ity of rea) storage
2F

2013 IBM Corporation

Information Management

More on !B2 10 an- 'ELE,SE4!E,LL5C,.E6


Prior re)eases: 'ELE,SE4!E,LL5C,.E6 not honore- 2hen pa 0age e3e $te- via !B,. 4i+e+& a !!H threa-6
/ Instea-& treate- as tho$gh (o$n- 2ith 'ELE,SE4C5MMI.6 / :hy: !B,.s an sti 0 aro$n- a L51# time& an- there 2as on ern that om(ination of 'ELE,SE4!E,LL5C,.E6 an- !B,.s 2o$)- ()o 0 $ti)ities& !!L& (in-s& et +

!B2 10: 2hen pa 0age (o$n- 2ith 'ELE,SE4!E,LL5C,.E6 is e3e $te- via a =reg$)ar> !B,.& that !B,. (e omes a high@ performan e !B,.
/ 'ELE,SE4!E,LL5C,.E6 honore/ Cigh@performan e !B,. -e-i ate- to onne tion thro$gh 2hi h it 2as instantiate-& vs+ going (a 0 into !!H threa- poo) 2hen transa tion en-s / If threa- re$se- for same pa 0age& yo$ get CP" (enefit of 'ELE,SE4!E,LL5C,.E6
2A 2013 IBM Corporation

Information Management

More on high@performan e !B,.s


Cigh@performan e !B,. 2i)) (e terminate- after (eing $seto pro ess 200 $nits of 2or0 4to free $p reso$r es6 Can s$spen- honoring of 'ELE,SE4!E,LL5C,.E6 for pa 0ages e3e $te- via !B2 (y iss$ing omman- @M5!IHJ !!H PI#'EL4C5MMI.6
/ Iss$e @M5!IHJ !!H PI#'EL4B1!5P.6 to s2it h (a 0

:hat if most of yo$r SLL e3e $te- thro$gh !!H is -ynami <
/ Consi-er (in-ing IBM !ata Server !river Pa 0age 4or !B2 Conne t6 pa 0ages into the -efa$)t 1"LLI! o))e tion 2ith 'ELE,SE4C5MMI.6& an- into another o))e tion 2ith 'ELE,SE4!E,LL5C,.E6
? Cave yo$r higher@vo)$me )ient@server transa tions $se that se ono))e tion to gain high@performan e !B,. performan e (enefits 4 o))e tion name an (e spe ifie- as a -ata so$r e property on the )ient si-e6
30 2013 IBM Corporation

Information Management

'ELE,SE4!E,LL5C,.E6 re ommen-ations
Best an-i-ates:
/ Pa 0ages freN$ent)y e3e $te-& an- e3e $te- via persistent threa-s s$ h as CICS@!B2 prote te- entry threa-s an- high@performan e !B,.s / Pa 0ages asso iate- 2ith (at h Bo(s that iss$e )ots of ommits
? Bat h (on$s: greater (enefit from -ynami prefet h& in-e3 )oo0asi-e

5perationa) onsi-erations:
/ Ens$re that 'ELE,SE4!E,LL5C,.E6 pa 0ages -o not a N$ire e3 )$sive ta()e spa e )o 0s 4 he 0 for )o 0 es a)ation& L5CI .,BLE6 / May nee- to restri t e3e $tion of 'ELE,SE4!E,LL5C,.E6 pa 0ages -$ring times of heavy $ti)ity e3e $tion& !!L& (in- operations& et +
? Bat h: ontro) thro$gh s he-$)ing ? !!H: $se @M5!IHJ !!H PI#'EL omman- 4see previo$s s)i-e6 ? CICS@!B2 prote te- threa-s: onsi-er hanging P'5.EC.1"M to 0 -$ring perio-s 2hen $ti)ities& pa 0age (in-8re(in-& an- !!L nee- to (e e3e $te31 2013 IBM Corporation

Information Management

5ther 2ays to tra-e memory for MIPS

32

2013 IBM Corporation

Information Management

!ynami statement a hing


#)o(a) statement a he has (een a(ove the 2 #B (ar in !BM1 a--ress spa e sin e !B2 KF
/ Size -etermine- (y QP,'M parameter E!MS.M.C

Larger statement a he O more a he =hits>


/ CP" savings a hieve- thro$gh res$)ting avoi-an e of f$)) P'EP,'Es

,t some point& yo$;- e3pe t to see -iminishing ret$rns from statement a he size in reases
/ If statement a he hit ratio is a)rea-y R A0M& f$rther a he size in reases may not yie)- signifi ant res$)ts 2ith respe t to CP" savings

33

2013 IBM Corporation

Information Management

'I! )ist pro essing


'I!s& or ro2 I!s& are ro2 )o ation in-i ators fo$n- in in-e3 entries !B2 pro esses 'I! )ists for things s$ h as:
/ List prefet h / In-e3 ,1!ing an- in-e3 5'ing / Cy(ri- Boin

CP" savings an (e a hieve- if 'I! poo) 4size- (y 2ay of QP,'M parameter M,P'BLI6 is )arge eno$gh to ena()e 'I! pro essing for a N$ery to omp)ete in memory
/ !B2 10: -efa$)t size of 'I! poo) 2ent to 700 MB& from F MB (efore
? 'I! poo) has (een a(ove the 2 #B (ar in !BM1 sin e !B2 KF ? .he )arger -efa$)t 'I! poo) size is one reason 2hy yo$ an e3pe t to see !B2 $ti)ization of rea) storage in rease 2hen yo$ migrate to !B2 10
37 2013 IBM Corporation

Information Management

In@memory 'I! )ist pro essing: a !B2 10 hange


:hat happens 2hen there is not eno$gh spa e in the 'I! poo) to a))o2 a 'I! )ist pro essing operation to omp)ete in memory:
/ Before !B2 10: !B2 2i)) a(an-on 'I! )ist pro essing an- go 2ith a ta()e spa e s an for -ata a ess / !B2 10: !B2 ontin$es 2or0ing on the 'I! )ist pro essing operation& $sing spa e in the 2or0 fi)e -ata(ase
? .hat;s (etter than giving $p an- fa))ing (a 0 to a ta()e spa e s an& ($t not N$ite as CP"@effi ient as getting entire 'I! )ist pro essing operation -one in memory ? CP" ost of omp)eting 'I! )ist pro essing $sing 2or0 fi)e spa e is re-$ eif the ($ffer poo) -e-i ate- to 32IB@page 2or0 fi)e ta()e spa es is )arge eno$gh to 0eep rea- I85 rate -o2n ? 1e2 QP,'M parameter M,P.EMPSS'I! an (e $se- to )imit amo$nt of 2or0 fi)e spa e that one 'I! )ist an $se
3% 2013 IBM Corporation

Information Management

.he sort poo)


.his is spa e in !BM1 4a(ove the 2 #B (ar sin e !B2 KF6 that is $se- for SLL@re)ate- sorts 4as oppose- to $ti)ity sorts6 .he )arger the poo)& the more CP"@effi ient SLL sorts 2i)) (e
/ Size -etermine- (y QP,'M parameter S'.P55L / 1ote that this is the ma3im$m size of the sort 2or0 area that !B2 2i)) a))o ate for ea h on $rrent sort $ser
? So& -on;t go over(oar- here if yo$ have a )ot of on $rrent SLL sort a tivity on yo$r system ? !efa$)t size of sort poo) 2ent to 10 MB 2ith !B2 10& from 2 MB (efore ? Ma3im$m S'.P55L va)$e is 12F MB / )argest I;ve seen on a !B2 system is 30 MB ? ,s 2ith any other memory@for@MIPS tra-e& if yo$ ma0e S'.P55L )arger& 0eep an eye on the z85S LP,';s -eman- paging rate / yo$ 2ant that to (e in the sing)e -igits 4or )ess6 per se on3D 2013 IBM Corporation

Information Management

.han0s for yo$r time


'o(ert Cattera)) rf atter*$s+i(m+ om

3E

2013 IBM Corporation

Das könnte Ihnen auch gefallen