ࡱ> `!OBCnJ\l3V"@v @Sx3xUk`ѴUu Ðl-(ZiC* Mpl"8e^/* _ xMvW&i6ż<ɓ_7 He  bYRa7n8S I~YHs7)a;*30)LLZ.C@j}ղYZgƪ/XpsoLghˉ;ُU L܀M5ų;?tWW׭ĺIj*%zLS50-^R3g9Y6f.-Y9˖ฎVWB 949ixv4::%*ۯD=Ӱ}>by2y2Q/'dbgl(zA V0͘#h ,(.^ UşbWER5 ؕztӛusqE՞bwBAHSIdR XLU۱lxlI!R-j.qi|=,kpwkxq %K2q}K|tFNXGsGb"9-N|kTM߽Cl 4؂gRo_Vg#ٞaW%)籓7(ґ[O7jKܿ^SQ_EA5RrOFLG:\.r@hhoWJ +PۯS"[])rEK&}[G$]o*\)28fnQHdQdC{qntԿ`L9,M5¿yhd@3*"a\Ž=hErm|!nGwUgzM_!ɑ JETMvv!>ϾMƣƙE.j4'DC; 4 p4yIhat)Z=kWiKMk4aB]+ksф> QgcȐgH ,h]t<#{F}@IBE UnuȐ{]'(vDWR#D*ʂsɐ/WGƉyR0)SǼ):{'<ŵ};w4gBVn΂w+ )H$PNG  IHDRPO5|gAMAPLTEech-0/`OآSpQ΄* U-i.ucKuA|reI⹑ˏװ϶Ъm'! pHYs.>NIDATxZqAEEI oIl{2=*e}~y 0_O-_s/TNCk5`.u5U&bŻq8`l@$5cc06͹8B"0?s4ZKenp\RXjO O+z*^J /{bK썍t5^|m拡 4dc#t'-;?zŋ.(v/vpHGB0kLrK7"DqT5Wk]Vmv)odᙡR4jaa[h}<- : *jUx@*&5nEf;DR@0:0J3hʻ0 o:lXڵ x2Ř p`z{gixzġ JPHD1?/*+ORfNj\"TXY'`8ć9×85^_P@D4 WmpxB&!^Ҕg%V7c"/EI2^W=4=JRdyKpEr &%coJ`pV早[J:|l)MW;](v4tn= F9s0߀Fޕ mz H9wDV#/߻9,\/E11` QKiCrpdR;(e=b$&3Z2{a{bzW?]֞1$"DN3o 0iц-wr/u0FB&;̗9 ;f<L]jQfzuQP";EWʅ6Su ~dqwy'^PW^\ 0bk0WI/<`KU\vXdl9fZY]x :aLxP(5O2e9Ym)ʅ23dAa-:,[@fe8štS'm$("W%W\99kow9X |Gtlw#Z9yϫ>m}IUbRsYQ\vI|Y,{z xW#\%2K@+I`*s"y+8|d`;M)Ћy:VΔС=oEi]ٛ#eS]Sߞf+6 ځ!j)W@oz^LIK<%s;G^u0XJ75\k&NDQz~Ytm.Y-;+X0a@N6As_P [T:BLݾEv#~ӷ`ݮ0V=8w\ʚ:凳С4A~6Ă擥טFdƒ4&[Ij: =g/:ԋ  !`"Cһ]re7Of, qqYD:Hע-ٮӟ;F.L kh9kaUPPUJ̝tZ`\:k(L!ȭ3^r!vՔD-jiP87z/9-<7eTr4`$hG$x|q)#x 4]](*o@ aeWLHw@ N޻Q5|/H8ͽBtXӵ)@sCDo~spf%p( 1)ih:/j_]aNRTiz^$J=ƺ|'+{N݆[ xD:~*$&=z`5cn+Jj4ڝIkg`$ Q]nL'TI6Eކ~P#MR!WLD_ù@Άi6ZN2B8ڄ_6>dm>!q 1C*mzBNʷTR#X{nށ'?D>;}m?AM'L[Ȳ(&G&k{#ڟ5_ q؃ C4$1(~,.äJLN;2j{QvբUx9ѿ:i%ML*__ q٣oU}v\;4El[i@ŧp~^?eܙn2#`9D0Sg }.H5;R䣝ׯ̀9nn+1]mˋw<-~=@ʎӘMe):(PMG52 8ryrQ\1mOY\NJ7\u_uxo>\ Bôx-Ūh?`dgG ${gH3Z ws.V[X_P--ljرVhbC=}g0vGAbM4 Fj{Uiy̞G[+#^ZϏBen,h;Ye1LVIENDB``!-V҈.\\x (ř(D9xXoF? ]dFCm0AT&&%Ci3Q,ԇ>lJ*q >q}˦{wg ivwο}Cq" ZR~}DcR;\Bf 0/U>A[/W3  +)|?0BD?q%i ľYH]حn=25G~]P!m+@A_c: ̅>mԇ=њ-xuo MhYMde||&2랗=υ{flXAÿ-FOОзKԖ,ͯO΀,8$L@prfs~wGYG90JE^6AEDYD4gk>6>baO!z?grA,d_d~>!Y>=q\zicE#m&LٱE6wABY'$9~uX6Հ:h"* ~80l gfh5MwOg>z{CmggHox /p厘zzl]Ȅ:Sg4IXPcjkrM^=r=x^"i_e Ђ rH0Ȯ})Yg&Z3h|W,=t?2 qrX{ؕ YH .xîLG0ozN58ԉ1T4xxlb+@> 'F}dc' Ec~FC9dSzi11y^.qIc',WѼyw r.YFpA>D}l2=0/~ljX6y9Ioiq9CNK)BEL{"Iq]8Slqeڅf{@{!GjfRsxk{=i"0qIQDfqhб6ck[:(OX_uY(0L=^ϖ0Sf S3m`s-8Cs9Cs33h![@shshK=K=Bhh< h![@s4 <[hбc[h^@b{ 4?5p/WGW%vf8Z G_{j"]E4%EɖPQ Bsts4.{{%B[q(K[.e-dKhn99:nBKBhix4"4niskFk%]dԸBXFl^]B8`p%Cgl#dl~xk},,qfE+:2!+<Ezޗ];e3' R~Mϯd=]YZnFGTG &UGgN;zꭙߺó'ߝ9})2@*Hv~/67 +GU\K]X_Q,[ dđ#6ydmOgZO1@X[ѯl4LlRDqd|g ^<>!\-coLJ,LVR1# !R1}!R̈́n@dŁݵz)e1LNզFpݩN&2LO}~C벂-GLeD^@&߽TWLWcZ[yUf(fFg/gglVќ2Pa# %>g{KffSveT#8\"Dk"X u]H7@ sVjG+w0<)\2Gs+=# ;408.XŒNb { Ft N;UމxGVJ.[CN.QjQWaHt߂ o+֒P @&ˠR|X$r| 9|p+QΙV1U6;!d5 %ϖ,]R(K8"<+_S_WЂFy]z3~gG+dcx2~|.*2cIL)5׭ՃqYaݧxUW}|%dDh00Dw'{z#l;`:+agr 7C~㘿|J,frdݝ W~F0[~ú*U;Ù\Azj[ و^W/^F}.4W|3nY[TO&SwRWA-CɗQ3yEE[Ù\e]$sv4AV1k~G:EXH9c5sk{̜Iz

aӲm1m1-ۦq=(Ebb/!l k[UAEoݚcx@,:[|-u9#[(:[S 82`$ S[ 8lh$z'xV!bpd~ u]w*w}rkkwZ(}"v$ɽjNajU SkU=~JS#Q@Bglouߎ}[սߤaIl/YfQ1|rNK&eUoPˏ&ˏںGv/B5rI`G0{PP$31V{+)iS+`x`D|Yaڹg|/F=fH{Z2>w:'iŐ`Di:h!oHΎd3,ӎޣY;4bOo䝦wwQ Mcif44|)a A X{T[4 jz!؈>?U>i:_ U[lv\uٌfk\ĈM#7cb s#cȝ\SS^i: #mKA2cmgvn:zdӀg݃=hz:t4V}S;R Udm,F ppL ƙEl_!K`!W|zl'K_VF)xKQǟ׳)"YJi F/AAD]" .G!_daW^Ώw͝wyjA"u pH//kw b/F3`vG::!IBŬ dئz%HȢ Qp BTv ֟Y*ח{K_q}dqR]%N羚XibZyg6흛eSUgVثثؖS&M['m2*#X" 5̄tG3!eofDUQ==sLS370s 1M&œ I\ 50#c!a*zόwyR|< _0rzf2IdR2޸e^+ɤaLj׊ibZEPNJZLGbp"mTӒ iUZ69,0~~V]qb:s Wi;|4 qL:|8Ӕsit6LSXwcrG N(349s#t9GSi=:u)ku1RNiOL6U)T6U)V%yDP4~f1Ǡsiʙ0"czO7;w=~SgG0+fŶ*K$k<ا_Êx69X#YGam)(Tل0<%=GsfE1x+pGĠa׊ :u > H;]yЗחu*-7V+^=ȏnT$~Fkuk;f.Axe_^3t 6]ǭxMۏHfls+<%v+^={nӵ}x]k= W(z    FZ01-ap1-n.ppt0-1,4,FCBJDocument Word.Document.80.Microsoft Word DocumentKDocument Word.Document.80.Microsoft Word DocumentDocument Word.Document.80.Microsoft Word DocumentDocument Word.Document.80.Microsoft Word DocumentDocument Word.Document.80.Microsoft Word Document/0(  0;[0 0 000$([\{b00 000000000  0=] 0 0 0000 2 3 !A0C0E0G0I0c00000000000000000!%),.:;?]}acdeghijklmnopDTimes New Roman(0(z[ 0 DArialNew Roman(0(z[ 0 " DWingdingsRoman(0(z[ 0 0DCourier Newman(0(z[ 0 1@DArial Blackman(0(z[ 0 "PDTimesNewRomanPS(0(z[ 0 `DHelvNewRomanPS(0(z[ 0 pDSymbolwRomanPS(0(z[ 0 DHelveticamanPS(0(z[ 0 DTimesicamanPS(0(z[ 0  A .@  @@`` @n?" dd@  @@`` D< ~l6G   !C Z /) C$ G%< 21~-D?#F+%//)) ! f !ff" 67?F>!< 5 #(;  CB;  ; !(= z-{:; }~EA #+ +%&(*/01-"; AA234567&% c!ZS   IJ=H00LMN'") $^_*YZTPgQR8LUgVsQEW8$EVK   *A>+P"#FG /2$BCnJ\l3V"@W$2$]n̖HB$Wb$΂w+ )H$V2$-V҈.\\x'$$$$$$$$$$$2$~9Y̖ =!2$W|zl'K_ -c $_ж_ж     A@  A1 8c8c     ?1 d0u0@Ty2 NP'p<'pA)BCD|E|| M dx<@8 m g4KdKd@z[ 0.pp<4ddddl 0CXg4dddd@z[ 0^p p<4!d!dl 0Cʚ;ʚ;<4ddddl|- 0X{9___PPT10 62___PPT9/ 0? -O =am2+Adabas Performance with Adabas v71 Features,+( , (Sierra Pacific Users Group June 11, 2002)( )  Introduction" (5  Old New Borrowed Bluex  M ADABAS v7.1" (5  KPerformance the old fashioned way Reduce I/O Reduce/Eliminate ADABAS CallsR"x*x!<" L + ADABAS v7.1" (5  Data Base Basics Associator - contains indices and control info Data Storage - contains compressed records Work - contains protection info and miscellaneous:xx  ( ADABAS v7.1" (5  7Back to the Basics ADABAS Calls Control Block 5 Buffers:x%x% 8 5ADABAS Architecture Overview  9ADABAS Architecture Overview  HAssociator Control Blocks for the DB/Files Inverted Lists of Descriptors& > > I ;@:ADABAS Architecture Overview   Data Storage  6ADABAS Architecture Overview  2Data Storage (continued) Records Files ISNs Fields& 3 ,ADABAS 7.1 - OVERHEAD  )ADABAS v7.1 - OVERHEAD"(5  ]Overhead per ADABAS call Systems Communication Buffer Movement Nucleus Processing Perhaps I/O:xExE ^ 8' ADABAS v7.1" (5  0 First Way to Reduce Overhead Reduce I/OH%x x$ < 1 R ADABAS v7.1" (5  BPerformance Forward Index Compression AND IT IS FREE ! ! !H x7x < C L%Adabas v71: Forward Index Compression6   M&Adabas v71: Forward Index Compression& KAnother example: Superdescriptor consisting of make, model and year (DL&  , Y&Adabas v71: Forward Index Compression& hFile Level ONLY! May reduce # Index levels FIC within single block only 1st value always stored in full &i Y i N &Adabas v71: Forward Index Compression& O &Adabas v71: Forward Index Compression& SSignificant space reduction Depends on descriptor specifics. Sample results below.287" 7 P &Adabas v71: Forward Index Compression& Q &Adabas v71: Forward Index Compression& R 'Adabas v71: Forward Index Compression ' YDO IT! any reasons not to??? No measurable CPU Set at file level Cannot be set with ONLINE REORDER Average space savings is 40-50% How to tell how much is saved? Run ADAORD 2x Use DDFILEA stats and ADAREP Fewer ASSO Reads/Writes Better buffer efficiency ADACSH; NI/UI blocks better candidates L3/L9/S1 commands are faster Fewer Index levels "$" !+ Mu#! !  *u Z X&Adabas v71: Forward Index Compression& Run usual ADAORD to eliminate the Unused-Rabn Chain and repack the blocks with padding factor Track the used NI/UI RABNS Run ADAORD again with Indexcompression on Note the new used NI/UI RABNS 2 runs of ADAORD - ugh!!!@*  b  =    \&Adabas v71: Forward Index Compression&    ~Index compression is done when the NI elements are being written FROM DDFILEA (and NOT to DDFILEA) So, look at the bytes used on DDFILEA and the resultant blocks written DDFILEA bytes = LRECL * ((EXCPBCKT - 1) /2)) After FIC completes, run ADAREP & add all used UI/NI blocks New bytes = Used-Asso-Rabns * Asso-Blocksize Savings percentage = (New bytes) / (DDFILEA bytes) Geeky way! J@$     @ S'Adabas v71: Forward Index Compression ' xCustomer results: University of Arkansas Production w/v712 In cyl. Before After Savings Percentage ADAPROD 3,200 1,400 1,800 56% ADASIS1 11,000 4,234 6,766 61% Cartridges required to backup the data base Before After ADAPROD 16 14 ADASIS1 45 37 ),d),d&?  7 T'Adabas v71: Forward Index Compression ' Customer results: University of Arkansas Production w/v712 ADADB1 Before After Commands 128,171,534 140,703,793 Asso Reads 1,777,290 1,539,641 Asso Writes 112,301 108,051 Buffer Eff. 48.0 52.0 ADADB2 Before After Commands 280,018,872 361,005,811 Asso Reads 11,173,438 6,503,484 Asso Writes 236,117 153,317 Buffer Eff. 54.0 70.0 T))  '  .  l  '  ,   W'Adabas v71: Forward Index Compression ' Customer results: Commonwealth of Mass: recovered 11,099 cylinders (58%) of Associator space due to the reorganization which implemented Forward Index Compression. Curtis Circulation: Normal and Upper Indices have been reduced typically by 50 to 80%. Files that were too large to cache before are now cacheable. University of Texas: w/ FIC Buffer efficiency 641.1 Asso Reads 5,560 w/o FIC Buffer efficiency 250.5 Asso Reads 14,268 ?g|f@s  /   j!Adabas v71: I/O Summary Overall Adabas CPU usage is reduced by an average of 10-20% Without FIC 5-10% reduction in I/Os to LBP With FIC 20-30% reduction in I/Os to LBP &   k"Adabas v71: User Results pQuotes from emails: 7-15% reduction in CPU with about a 5% reduction in ASSO I/O when doing Read Logical and Finds. This is related to the improvements to the improvement in how the buffer chooses the replacement victims. (Not using FIC) With version 622 we used 17:54 CPU hours to process 119 million commands and with 712 we used 12:40 to process 115 million. I also noticed that the IOSQ time of our DASD went down dramatically. I have 7 production databases. CPU savings from v712 are: 7.6%, 8.7%, 8.9%, 10.8%, 14.7%, 18.6%and 0.01% We have an average of 30% CPU savings on each of 2 databases converted to v71 @]X q l#Adabas v71: More User Results MCustomer states: Database 233 (LBP=360 MB) : V62: 0.43 milliseconds CPU per call or 2334 calls per CPU second internal throughput V71: 0.26 milliseconds CPU per call or 3776 calls per CPU second internal throughput This is a reduction in CPU time of more than 38% and an increase in internal throughput of nearly 62% (!) P;: N m$Adabas v71: More User Results GSame Customer: Database 15 (LBP=360 MB) : V62: 0.34 milliseconds CPU per call or 2930 calls per CPU second internal throughput V71: 0.23 milliseconds CPU per call or 4289 calls per CPU second internal throughput This is a reduction in CPU time of more than 31% and an increase in internal throughput of more than 46%(9G H "Adabas Buffer Pool  JHolds Associator and Data blocks Size Set by DBA Larger Buffer = Less I/O  K %Look Ahead : 64-bit Overview - V7.1.3 & 64-bit Real Storage OS/390 V2R10 and z/OS 1.1 Max. 128 GB Real Storage (temporarily) 2 GB size limitation for allocation 64-bit Virtual Storage z/OS 1.2 2 GB size limitation removed data caches only other structures in later versions of z/OS^eb3 %b  OI/O Buffer Pool (LBP)  Descriptor usage Use fewer descriptors in search Re-order file by most-used descriptor Typical size Small DB: 5MB Medium DB: 10-20MB Large DB: 30-50MB Very Large DB: 70+MBTF HF H  NASSO/DATA Block Sizes  Various block sizes waste various amounts of LBP? LBP is divided into 256 byte slots Assume 3390 device: ASSO block size 2544 2544 / 256 = 10 slots 1,000,000 / 2544 = 394 * 10 = 3940 * 256 = 1,008,640 bytes in LBP Assume 8392 device: ASSO block size 4092 4092 / 256 = 16 slots 1,000,000 / 4092 = 245 * 16 = 3920 * 256 = 1,003,520 bytes in LBP Conclusion Larger block sizes save very little in the buffer pool EXERCISE FOR THE USER !2LX)X]2LX)X R  PAdabas Caching Facility  A buffer in addition to the Adabas buffer pool -  Read Only For selectable data User-specific - RABNs User specified file(s) WORK parts 2 and 3 Cache buffer location Extended memory Data space Hiperspace (depending on operating system) CBUFNO=n Read ahead to end of track Affects L2/L5, L3/L6, L9, and S1 (non-DE search)TAF LTAFL  &e   & Advantages  BLarger Cache results in less I/Os More space in the 31-bit area  C S ADABAS v7.1" (5  C Second Way to Reduce Overhead Reduce/Eliminate ADABAS CallsH&xx%< D 0 ADABAS v7.1" (5  ZHOW TO REDUCE ADABAS CALLS FastPath Prefetch / Multifetch Change Programs Add New Keys :x@x@ [ - ADABAS v7.1" (5  HOW TO REDUCE ADABAS CALLS W I T H O U T CHANGING PROGRAMS ! FastPath - A data buffering approach Prefetch / Multifetch - A read ahead approach :@xgx?g  3ADABAS v7.1 - FastPath"(5  FastPath uses its own buffer to hold data and control information. FastPath can be monitored on-line Parameters control what FastPath optimizes 4xx  4ADABAS v7.1 - FastPath"(5  FastPath has many displays available to show what is being optimized and the savings Some parameters can be modified on-line while FastPath is active 4xx  2Adabas Fastpath Statistics4     EFASTPATH - normal look up    FD&Adabas Fastpath Statistics - Aha !4    GFASTPATH - WHY ?   JADABAS facilities   AFastpath Hints  1How to determine which files to Fastpath. Determine top ten files per physical Database. Fastpath ten files - but watch out for update activity. Monitor for week to make sure resources are adequate. Add files until all resources consumed or there are no more high volume files to fastpath F*-*Z   A     CFastpath Hints  Reasons Commands may be excluded from a File Insufficient CID/User Areas. ADABAS / File processing is stopped. A Job step is running that has "Prefetch" parameters. The "Command / Descriptor" is not defined in the file Parameters Record Buffer Length / Set ID Length Exceeded Set ID Length = Search Buffer + Format Buffer field length values + other factors. Minimum Sequential Storage (k) exceeded. FAPP,&   B%!ADABAS PREFETCH/MULTIFETCH- Batch " kUses a Read Ahead approach Needs a Buffer on the User Side Controlled by ADARUN parameters Very Easy to Usel 5wlQ ?!ADABAS PREFETCH/MULTIFETCH- Batch " PREFETCH=OLD for Prefetch PREFETCH=NEW for Multifetch PREFSBL=nnnn for setting work area size PREFXFIL=nn and PREFXCMD used in pairs to exclude a file/command from Prefetch 5w =!ADABAS PREFETCH/MULTIFETCH- Batch " PREFETCH Brings Back ALL of the Prefetched Data in the ISN Buffer - with a 16 byte header for each Record returned MULTIFETCH Brings back the Records in the Record Buffer without any separation - The 16 byte headers are returned in the ISN Buffer 5w >"ADABAS PREFETCH/MULTIFETCH- Online # VERY SPECIAL CASE VERY DETAILED CODING INVOLVED BYTE SHIFTING/MOVING REQUIRED BUT CAN BE VERY REWARDING In NATURAL you can use the USR1043N ENTRY POINT for ADABAS Direct Calls !  5w Q1ALTERNATE IDEA for reducing calls to ADABAS !21 2 eTake it OFF LINE ! ADASTRIP NATURAL Program Other ways GREAT for NATURAL WORK Files / SORTs>$)$)&`   V!ADASTRIP Product Overview - 1""! " Fast, efficient extraction of multiple data files from one or more ADABAS files One pass through backup tapes or database to extract data needed Flat files can be input to pre-processing programs or loaded directly to the database User-exits available for extra processing $Z  W!ADASTRIP Product Overview - 2""! " "Implement ADASTRIP to extract data from the ADASAV backup tapes All data can be extracted in one run No impact on on-line databases - OFF LINE ! Most recent data can be merged from the latest QDUMP or PLOG Extra processing of data can be done after extraction prior to load of database#$Z# # X Operations,   Y!ADASTRIP Product Overview - 4""! " Extraction of the file structure (FDT) only Normalization of PE/MUs is possible Recovery of historical data from past backups Creation of test data Deferred processing of critical reports$Z&@  y Z!ADASTRIP Product Overview - 5""! " Recovery of historical data Recovery of  lost NATURAL programs Export to other systems or DBMSs Import data between ADABAS databases Fast processing avoids ADABAS I/O Deferred processing Replace ADAULD$Z&[  k [!ADASTRIP Product Overview - 6""! " Moving sequential processing away from ADABAS reduces load Faster processing, avoids ADABAS I/O Deferred processing, reports can be run from data extracted from a backup Processing on synchronized data (a backup has the data fixed at a point in time)$Z  U Monitoring   CPU utilization ADANC7: Format translations ADANC2: Command execution (L1, etc.) ADANC8: Compression/decompression Monitoring packages Provide valuable information Can use considerable CPU cycles and I/OTcEcE  <Converting to V7  What Gets Converted GCB Version; XGCBSTAT=x C7 DSECT changes FCB DSECT changes new fields (YIVDE initialized) FST Blocks 10-14 Blocks 11-14 formatted WORK formatted PLOG formattedN 5w,3$ TConverting to V7  New SVC Will process both V7 and V6 data bases Old V6 SVC will work for basic V7  stuff Not for TCP/IP, UES, AFP & u 5wu0E & THE END  Until Next Time !  /p |),0V|}~PD ` ̙33` ` ff3333f` 333MMM` f` f` 3` >?" dZ R?ndRdlPIndX%  P5 r?" dd@%   @@``@n?" dd@  @@``PP   @ ` `0p>>   R (  |  TZ?? 9 |  TZ??{ |  TZ??c K  Zxaxa1 ?  ]Firstlevel: Arial bold, 24 point Second level: Arial, 24 point Third level: Arial, 20 point ! ^    `xaxa1 ?0  a-Headline, Arial bold, 32 point 2 lines max. . .^B  6DR?|  TZ?M? iF P   P   # lZ?Md1?``   0?P`   0R?`   Ht? ?0~  *.(l2    8Xh\  S ~_ж_ж 1?31 dSierra Pacific Users Group Meeting  June 11, 20023 23  3 B  s *޽h ? R  sag_template.ppt  e]`(     `TKxaxa1 ? K X Click to edit Master title style!! ~F  9      TZ?? 9   TZ?? {  TZ?? c  TZ?M? i^B  6DR?   HK1 ?H   K [#Click to edit Master subtitle style$$ F P   P   # lZ?Md1?``   0?P`   0R?`  HK? ?0~ K *.(l2    8XhB  s *޽h ? P 0 @`(     # l 8c?  O X*    # l 8c? z  Z*  v  6 ?;@  V  # lPĥ 8c?x U%b  RClick to edit Master text styles Second level Third level Fourth level Fifth level!    S    # l 8c?    X*    # lT 8c? z   Z*  H  0xl>L ? ̙3380___PPT10.Ff  0 0(   B  s *xl>L ? a(80___PPT10.Fa}h  Pd$( p  d  d ZKxaxa1 ?` P tDavid Del Rio (_ r  d S K  K l d C ,K pP  K H d 0޽h ? ̙33  | J(  x  c $XO0  O   S 5O  O ,@h8H  0޽h ? A|o^U    p J(  p x p c $=O0  O  p S `BO  O ,@h8H p 0޽h ? A|o^U    x J(  x x x c $NO0  O  x S TO  O ,@h8H x 0޽h ? A|o^U    ` J(  ` x ` c $ZO0  O  ` S bO  O ,@h8H ` 0޽h ? A|o^U  K0 d\&& (     C xgOxaxa1 ?0  O   ZhiOxaxa1?@P p iADABAS Nucleus *  XB  01? E\ XB  01? U <XB  01?e <XB  01?@@XB  01?U <6 ^  61? @ ^  61? ^  61?@ ` ^2  61? p ^  61? p ^  61?  ^2  61? 0 ^2  61?p p ^  61? p ^  61?`  ^  61?  ^2  61?  ^2  61?0  ^2  61? p0 ^  61? ` P ^2  61? ` P ^  61?@ ` ^2  61? p ^  61? P @ ^2  61?p P @ ^  61? p ^2  61? p ^  61?0 P ^2  61? `  ! ZxqOxaxa1?4 iZ X Data Storage      " TuOxaxa1?   T Associator     # ZyOxaxa1?-  NWork    $ Z}Oxaxa1? * ZProtection Log    % Z<Oxaxa1?}  f Command Log *       & ZOxaxa1?@P 0  eI/O Buffer Pool&  H  0޽h ? f3f3M  K0   u(     C xOxaxa1 ?0  O   C x8Oxaxa1 ? O   ZOxaxa1? P `  dInverted Lists for Files    ZOxaxa1?P @  XControl Blocks  ^B  61?  ^B  61? \   ZOxaxa1? P   wAddress Converters for Files *  ^B  61?P  P ^  61?  ^2  61?`  ^  61? `  ^2  61? `    TĝOxaxa1? -S  ^ Associator "    H  0޽h ? f3f3&K0 @ N(     ZOxaxa1? fADABAS Architecture Overview,  ~  TЬOxaxa1?@` JAssociator (continued) Control Blocks for Files File Control Block (FCB) j%        K ^  61?  ^2  61?`  ^  61? `  ^2  61? `    TOxaxa1? -S  ^ Associator "      ZOxaxa̙1?   Y FCB - File 16  ^B  61?@ U@   Z|Oxaxa̙1?  p Y FCB - File 17  ^B  61? U   ZOxaxa1? s   G. .  ^B  61?@ U @ ^B  61? U  ^  61? ``  ZOxaxa1? ` oRABN RABN RABN RABN "    XB  01? XB  01? 00XB  01? ^  61? `0   ZOxaxa1? `  r*Field Field Field Field++  + XB  01?  XB  01? 00 XB  01?    Z`Oxaxa1? -  QFile 17    Z  _Inverted List - Last-Name  ^ % 61? P ^2 & 61?` P ^ ' 61? ` ^2 ( 61? `  ) T4Qxaxa1? m  ^ Associator "    H  0޽h ? f3f3M  K0    u (     C x9Qxaxa1 ?0  Q   C x:Qxaxa1 ? Q ^  6f1?L < ^2  61? @x ^  6f1? xl^2  61? x   Z+B#style.visibility<* l %(DA' =%(D' =%(D' =A@BBBB0B%(D' =1:Bvisible*o3>+B#style.visibility<* l 5%(DA' =%(D' =%(D' =A@BBBB0B%(D' =1:Bvisible*o3>+B#style.visibility<* l 5i%(DA' =%(D' =%(D' =A@BBBB0B%(D' =1:Bvisible*o3>+B#style.visibility<* l i%(DA' =%(D' =%(D' =A@BBBB0B%(D' =1:Bvisible*o3>+B#style.visibility<* l %(DA' =%(D' =%(D' =A@BBBB0B%(D' =1:Bvisible*o3>+B#style.visibility<* l %(+8+0+ l U +q   !  ` t (  t x t c $D)UP  U r t S *U!  U   t <`+U W  Standard Compression 104 bytes6!  ^  t 6? l   t <1U    T 26 CHYRSLERTOWNANDCOUNTRY1998 26 CHYRSLERTOWNANDCOUNTRY1999 26 CHYRSLERTOWNANDCOUNTRY2000 26 CHYRSLERTOWNANDCOUNTRY2001 :          ^ t 6? l  t <>U   ;q 26 0 CHYRSLERTOWNANDCOUNTRY1998 2 25 9 5 22 2000 2 25 1 :r  ,       g XB t 0D  l XB t 0D l  t <hKU p 0Forward Index Compression 40 bytes 61% savings61"XB t 0Dl H t 0޽h ? ̙33  p *(   x  c $WU0  U r  S XU! U H  0޽h ? ̙33  x 0(  x x x c $^U0  U x x  <A ??(8 H x 0޽h ? ̙33  bZ| (   | R | s *kx | c $ cU0  U ` |  c $A ??E x | c $ 2) (S1, S2, S8, S9) FASTPATH uses Read Ahead ! !v 0Z<0(Z0Z0Z <    H 8 0޽h ? 6  @, v( / , x , c $Y0  Y  ,  BA ?8c?` .XL 8H , 0޽h ? A|o^U v K0 Pf@ (  @  @ 3 rP] jJ jJ ?Pf\  ]  f@ N(]gֳgֳ?e V 80% of accesses to 20% data No need to  preload entire files Challenge is to identify  popular data in  popular formats Constantly re-identify popular data0(Z0Z$0(Z0Z)0(Z0Z%0(Z  H @ 0޽h ? P K0 p77Y x( @ X  X 3 rd] jJ jJ ?pf/   ] F   X    X  C B C DEF 1?AA=]{{]=]6q{X]C=/\6/CXrqXC/6]/=C]X{r6\@ b X   2BCDEDFL1?'N|sreUA+,BUert|N'#$@< b X   2BCDEDFL1?]|6reUB,qXC/sN'#$@ = b X   2BCDEDFL1?\6/CXqqXC/6]#$@ = b X   2BCDEDFL1?'Nt/CXq+AUer6|\#$@<   X  pBCDEF ?AA|T/ oWlAO.1}V0 .AWopWA. /U|0.PAmWp 0V}1OlmO0|@   X  pBCDEF ?AAuO*m|Uc@G-) zT/ -@UnoWB.*Ou'.FBbW|o /Tz )Gc||bE'u@   X  pBCDEF ?AAoI%lqTX?=- uP+-?UloVA.%Jo.<AXVro+Pu =XrqX;o@   X  pBCDEF ?AAhC|ifRO>5+qM(+>RjoVA.!Ei.2ANVgo~(Mq5~OgfM|1h@   X  pBCDEF ?AAb=pg\QE<++nI&+=QhmU@->c -(@CU\ms&In+rE\\Bp' b@