bdb_usenix.ps

介绍：MySQL是比较出名的数据库软件

79.2 595.2 R 3.139(yr)-.15 G .639
(ead operation will need to visit the disk)-3.139 F 1.201
(in the steady state.)79.2 607.2 R 1.201
(The programmer declares the size)6.201 F(of the cache re)79.2 619.2 Q
(gion at startup.)-.15 E(Finally)79.2 635.4 Q 7.048(,m)-.65 G(an)-7.048
E 7.048(yo)-.15 G 4.548(perating systems pro)-7.048 F 4.548
(vide memory-)-.15 F 2.532(mapped \214le services that are much f)79.2
647.4 R 2.533(aster than their)-.1 F 2.602
(general-purpose \214le system interf)79.2 659.4 R 5.102(aces. Berk)-.1
F(ele)-.1 E 5.102(yD)-.15 G(B)-5.102 E 5.118
(can memory-map its database \214les for read-only)79.2 671.4 R 3.917
(database use.)79.2 683.4 R 3.917(The application operates on records)
8.917 F 2.069(stored directly on the pages, with no cache manage-)79.2
695.4 R 1.557(ment o)79.2 707.4 R -.15(ve)-.15 G 4.057(rhead. Because)
.15 F 1.556(the application gets pointers)4.057 F 1.265
(directly into the Berk)323.2 84 R(ele)-.1 E 3.765(yD)-.15 G 3.765(Bp)
-3.765 G 1.265(ages, writes cannot be)-3.765 F 3.775
(permitted. Otherwise,)323.2 96 R 1.275(changes could bypass the lock-)
3.775 F .23(ing and logging systems, and softw)323.2 108 R .23
(are errors could cor)-.1 F(-)-.2 E 4.007(rupt the database.)323.2 120 R
4.006(Read-only applications can use)9.007 F(Berk)323.2 132 Q(ele)-.1 E
2.893(yD)-.15 G(B')-2.893 E 2.893(sm)-.55 G .393
(emory-mapped \214le service to impro)-2.893 F -.15(ve)-.15 G
(performance on most architectures.)323.2 144 Q F1 3
(3.6. Con\214gurable)323.2 174 R(page size)3 E F0 .111
(Programmers declare the size of the pages used by their)323.2 190.2 R
.403(access methods when the)323.2 202.2 R 2.903(yc)-.15 G .403
(reate a database.)-2.903 F(Although)5.403 E(Berk)323.2 214.2 Q(ele)-.1
E 4.046(yD)-.15 G 4.046(Bp)-4.046 G(ro)-4.046 E 1.546
(vides reasonable def)-.15 F 1.546(aults, de)-.1 F -.15(ve)-.25 G
(lopers).15 E 3.64(may o)323.2 226.2 R -.15(ve)-.15 G 3.64
(rride them to control system performance.).15 F .793
(Small pages reduce the number of records that \214t on a)323.2 238.2 R
.353(single page.)323.2 250.2 R(Fe)5.353 E .353
(wer records on a page means that fe)-.25 F(wer)-.25 E .724
(records are lock)323.2 262.2 R .724(ed when the page is lock)-.1 F .723
(ed, impro)-.1 F(ving)-.15 E(concurrenc)323.2 274.2 Q 5.262 -.65(y. T)
-.15 H 1.462(he per).65 F 1.462(-page o)-.2 F -.15(ve)-.15 G 1.462
(rhead is proportionally).15 F 2.29
(higher with smaller pages, of course, b)323.2 286.2 R 2.29(ut de)-.2 F
-.15(ve)-.25 G(lopers).15 E(can trade of)323.2 298.2 Q 2.5(fs)-.25 G
(pace for time as an application requires.)-2.5 E F1 3(3.7. Small)323.2
328.2 R -.3(fo)3 G(otprint).3 E F0(Berk)323.2 344.4 Q(ele)-.1 E 3.973
(yD)-.15 G 3.973(Bi)-3.973 G 3.974(sac)-3.973 G 1.474(ompact system.)
-3.974 F 1.474(The full package,)6.474 F .832
(including all access methods, reco)323.2 356.4 R -.15(ve)-.15 G
(rability).15 E 3.331(,a)-.65 G .831(nd trans-)-3.331 F 1.235
(action support is roughly 175K of te)323.2 368.4 R 1.236
(xt space on com-)-.15 F(mon architectures.)323.2 380.4 Q F1 3
(3.8. Cursors)323.2 410.4 R F0 1.57(In database terminology)323.2 426.6
R 4.07(,ac)-.65 G 1.57(ursor is a pointer into an)-4.07 F 1.806
(access method that can be called iterati)323.2 438.6 R -.15(ve)-.25 G
1.807(ly to return).15 F 3.68(records in sequence.)323.2 450.6 R(Berk)
8.68 E(ele)-.1 E 6.18(yD)-.15 G 6.18(Bi)-6.18 G 3.68(ncludes cursor)
-6.18 F(interf)323.2 462.6 Q 2.814(aces for all access methods.)-.1 F
2.815(This permits, for)7.814 F -.15(ex)323.2 474.6 S .34
(ample, users to tra).15 F -.15(ve)-.2 G .34(rse a B+tree and vie).15 F
2.84(wr)-.25 G .34(ecords in)-2.84 F(order)323.2 486.6 Q 6.233(.P)-.55 G
1.234(ointers to records in cursors are persistent, so)-6.233 F 1.779
(that once fetched, a record may be updated in place.)323.2 498.6 R
(Finally)323.2 510.6 Q 4.438(,c)-.65 G 1.939
(ursors support access to chains of duplicate)-4.438 F
(data items in the v)323.2 522.6 Q(arious access methods.)-.25 E F1 3
(3.9. J)323.2 552.6 R(oins)-.18 E F0 2.703(In database terminology)323.2
568.8 R 5.203(,aj)-.65 G 2.702(oin is an operation that)-5.203 F .616
(spans multiple separate tables \(or in the case of Berk)323.2 580.8 R
(e-)-.1 E(le)323.2 592.8 Q 4.518(yD)-.15 G 2.018
(B, multiple separate DB \214les\).)-4.518 F -.15(Fo)7.017 G 4.517(re)
.15 G 2.017(xample, a)-4.667 F(compan)323.2 604.8 Q 3.372(ym)-.15 G .873
(ay store information about its customers in)-3.372 F 1.545
(one table and information about sales in another)323.2 616.8 R 6.545
(.A)-.55 G(n)-6.545 E 1.498(application will lik)323.2 628.8 R 1.499
(ely w)-.1 F 1.499(ant to look up sales informa-)-.1 F .933
(tion by customer name; this requires matching records)323.2 640.8 R
2.28(in the tw)323.2 652.8 R 4.78(ot)-.1 G 2.28
(ables that share a common customer ID)-4.78 F 2.515(\214eld. This)323.2
664.8 R .015(combining of records from multiple tables is)2.515 F
(called a join.)323.2 676.8 Q(Berk)323.2 693 Q(ele)-.1 E 5.561(yD)-.15 G
5.561(Bi)-5.561 G 3.061(ncludes interf)-5.561 F 3.062
(aces for joining tw)-.1 F 5.562(oo)-.1 G(r)-5.562 E(more tables.)323.2
705 Q EP
%%Page: 5 5
%%BeginPageSetup
BP
%%EndPageSetup
/F0 12/Times-Bold@0 SF 3(3.10. T)79.2 84 R(ransactions)-.888 E/F1 10
/Times-Roman@0 SF -.35(Tr)79.2 100.2 S(ansactions ha).35 E .3 -.15(ve f)
-.2 H(our properties [Gray93]:).15 E/F2 8/Times-Roman@0 SF<83>84.2 116.4
Q F1(The)17.2 E 5.489(ya)-.15 G 2.989(re atomic.)-5.489 F 2.989
(That is, all of the changes)7.989 F 1.475
(made in a single transaction must be applied at)104.2 128.4 R 1.31
(the same instant or not at all.)104.2 140.4 R 1.31(This permits, for)
6.31 F -.15(ex)104.2 152.4 S 3.565(ample, the transfer of mone).15 F
6.065(yb)-.15 G 3.565(etween tw)-6.065 F(o)-.1 E 3.68
(accounts to be accomplished, by making the)104.2 164.4 R 1.27
(reduction of the balance in one account and the)104.2 176.4 R
(increase in the other into a single, atomic action.)104.2 188.4 Q F2
<83>84.2 204.6 Q F1(The)17.2 E 3.125(ym)-.15 G .625(ust be consistent.)
-3.125 F .625(That is, changes to the)5.625 F 3.628(database by an)104.2
216.6 R 6.128(yt)-.15 G 3.628(ransaction cannot lea)-6.128 F 3.929 -.15
(ve t)-.2 H(he).15 E(database in an ille)104.2 228.6 Q -.05(ga)-.15 G
2.5(lo).05 G 2.5(rc)-2.5 G(orrupt state.)-2.5 E F2<83>84.2 244.8 Q F1
(The)17.2 E 3.006(ym)-.15 G .506(ust be isolatable.)-3.006 F(Re)5.506 E
-.05(ga)-.15 G .505(rdless of the num-).05 F .8(ber of users w)104.2
256.8 R .8(orking in the database at the same)-.1 F 1.88(time, e)104.2
268.8 R -.15(ve)-.25 G 1.88(ry user must ha).15 F 2.18 -.15(ve t)-.2 H
1.88(he illusion that no).15 F(other acti)104.2 280.8 Q
(vity is going on.)-.25 E F2<83>84.2 297 Q F1(The)17.2 E 5.54(ym)-.15 G
3.04(ust be durable.)-5.54 F(Ev)8.04 E 3.04(en if the disk that)-.15 F
.877(stores the database is lost, it must be possible to)104.2 309 R
(reco)104.2 321 Q -.15(ve)-.15 G 2.668(rt).15 G .168
(he database to its last transaction-consis-)-2.668 F(tent state.)104.2
333 Q 2.49(This combination of properties \212 atomicity)79.2 349.2 R
4.99(,c)-.65 G(onsis-)-4.99 E(tenc)79.2 361.2 Q 4.542 -.65(y, i)-.15 H
3.243(solation, and durability \212 is referred to as).65 F -.4(AC)79.2
373.2 S 3.459(IDity in the literature.).4 F(Berk)8.459 E(ele)-.1 E 5.958
(yD)-.15 G 3.458(B, lik)-5.958 F 5.958(em)-.1 G(ost)-5.958 E .993
(database systems, pro)79.2 385.2 R .993(vides A)-.15 F .994
(CIDity using a collection)-.4 F(of core services.)79.2 397.2 Q .257
(Programmers can choose to use Berk)79.2 413.4 R(ele)-.1 E 2.757(yD)-.15
G(B')-2.757 E 2.757(st)-.55 G(ransac-)-2.757 E
(tion services for applications that need them.)79.2 425.4 Q F0 3
(3.10.1. Write-ahead)79.2 455.4 R(logging)3 E F1 .479
(Programmers can enable the logging system when the)79.2 471.6 R(y)-.15
E .918(start up Berk)79.2 483.6 R(ele)-.1 E 3.418(yD)-.15 G 3.418
(B. During)-3.418 F 3.417(at)3.417 G .917(ransaction, the appli-)-3.417
F .493(cation mak)79.2 495.6 R .493
(es a series of changes to the database.)-.1 F(Each)5.494 E .552
(change is captured in a log entry)79.2 507.6 R 3.052(,w)-.65 G .552
(hich holds the state)-3.052 F .207
(of the database record both before and after the change.)79.2 519.6 R
2.208(The log record is guaranteed to be \215ushed to stable)79.2 531.6
R .871(storage before an)79.2 543.6 R 3.371(yo)-.15 G 3.371(ft)-3.371 G
.871(he changed data pages are writ-)-3.371 F 3.989(ten. This)79.2 555.6
R(beha)3.989 E 1.489(vior \212 writing the log before the data)-.2 F
(pages \212 is called)79.2 567.6 Q/F3 10/Times-Italic@0 SF
(write-ahead lo)2.5 E -.1(gg)-.1 G(ing).1 E F1(.)A .835(At an)79.2 583.8
R 3.335(yt)-.15 G .835(ime during the transaction, the application can)
-3.335 F F3(commit)79.2 595.8 Q F1 4.202(,m)C 1.702
(aking the changes permanent, or)-4.202 F F3 -.45(ro)4.201 G 1.701
(ll bac).45 F(k)-.2 E F1(,)A .852
(cancelling all changes and restoring the database to its)79.2 607.8 R
1.57(pre-transaction state.)79.2 619.8 R 1.57
(If the application rolls back the)6.57 F 1.003
(transaction, then the log holds the state of all changed)79.2 631.8 R
.5(pages prior to the transaction, and Berk)79.2 643.8 R(ele)-.1 E 3(yD)
-.15 G 3(Bs)-3 G(imply)-3 E .226(restores that state.)79.2 655.8 R .226
(If the application commits the trans-)5.226 F .538(action, Berk)79.2
667.8 R(ele)-.1 E 3.038(yD)-.15 G 3.038(Bw)-3.038 G .538
(rites the log records to disk.)-3.038 F(In-)5.537 E 2.312
(memory copies of the data pages already re\215ect the)79.2 679.8 R
1.399(changes, and will be \215ushed as necessary during nor)79.2 691.8
R(-)-.2 E 2.35(mal processing.)79.2 703.8 R 2.35
(Since log writes are sequential, b)7.35 F(ut)-.2 E 8.732
(data page writes are random, this impro)79.2 715.8 R -.15(ve)-.15 G(s)
.15 E(performance.)323.2 84 Q F0 3(3.10.2. Crashes)323.2 114 R(and r)3 E
(eco)-.216 E -.12(ve)-.12 G(ry).12 E F1(Berk)323.2 130.2 Q(ele)-.1 E
3.592(yD)-.15 G(B')-3.592 E 3.592(sw)-.55 G 1.093
(rite-ahead log is used by the transac-)-3.592 F .415
(tion system to commit or roll back transactions.)323.2 142.2 R .414
(It also)5.414 F(gi)323.2 154.2 Q -.15(ve)-.25 G 3.23(st).15 G .73
(he reco)-3.23 F -.15(ve)-.15 G .73
(ry system the information that it needs).15 F .824(to protect ag)323.2
166.2 R .824(ainst data loss or corruption from crashes.)-.05 F(Berk)
323.2 178.2 Q(ele)-.1 E 2.703(yD)-.15 G 2.703(Bi)-2.703 G 2.704(sa)
-2.703 G .204(ble to survi)-2.704 F .504 -.15(ve a)-.25 H .204
(pplication crashes, sys-).15 F .408(tem crashes, and e)323.2 190.2 R
-.15(ve)-.25 G 2.908(nc).15 G .407(atastrophic f)-2.908 F .407
(ailures lik)-.1 F 2.907(et)-.1 G .407(he loss)-2.907 F
(of a hard disk, without losing an)323.2 202.2 Q 2.5(yd)-.15 G(ata.)-2.5
E(Survi)323.2 218.4 Q .538(ving crashes requires data stored in se)-.25
F -.15(ve)-.25 G .539(ral dif).15 F(fer)-.25 E(-)-.2 E 2.52(ent places.)
323.2 230.4 R 2.52(During normal processing, Berk)7.52 F(ele)-.1 E 5.02
(yD)-.15 G(B)-5.02 E .766(has copies of acti)323.2 242.4 R 1.066 -.15
(ve l)-.25 H .766(og records and recently-used data).15 F 1.539
(pages in memory)323.2 254.4 R 6.539(.L)-.65 G 1.539
(og records are \215ushed to the log)-6.539 F .694
(disk when transactions commit.)323.2 266.4 R .695
(Data pages trickle out)5.694 F .008(to the data disk as pages mo)323.2
278.4 R .308 -.15(ve t)-.15 H .008(hrough the b).15 F(uf)-.2 E .008
(fer cache.)-.25 F(Periodically)323.2 290.4 Q 2.691(,t)-.65 G .191
(he system administrator backs up the data)-2.691 F .278
(disk, creating a safe cop)323.2 302.4 R 2.778(yo)-.1 G 2.778(ft)-2.778
G .278(he database at a particular)-2.778 F 2.609(instant. When)323.2
314.4 R .109(the database is back)2.609 F .109(ed up, the log can be)-.1
F 3.838(truncated. F)323.2 326.4 R 1.337(or maximum rob)-.15 F 1.337
(ustness, the log disk and)-.2 F(data disk should be separate de)323.2
338.4 Q(vices.)-.25 E(Dif)323.2 354.6 Q 1.29(ferent system f)-.25 F 1.29
(ailures can destro)-.1 F 3.79(ym)-.1 G(emory)-3.79 E 3.79(,t)-.65 G
1.29(he log)-3.79 F 1.106(disk, or the data disk.)323.2 366.6 R(Berk)
6.106 E(ele)-.1 E 3.606(yD)-.15 G 3.606(Bi)-3.606 G 3.606(sa)-3.606 G
1.106(ble to survi)-3.606 F -.15(ve)-.25 G .679(the loss of an)323.2
378.6 R 3.179(yo)-.15 G .679(ne of these repositories without losing)
-3.179 F(an)323.2 390.6 Q 2.5(yc)-.15 G(ommitted transactions.)-2.5 E
1.372(If the computer')323.2 406.8 R 3.871(sm)-.55 G 1.371
(emory is lost, through an applica-)-3.871 F 1.619
(tion or operating system crash, then the log holds all)323.2 418.8 R
1.789(committed transactions.)323.2 430.8 R 1.788(On restart, the reco)
6.789 F -.15(ve)-.15 G 1.788(ry sys-).15 F .49(tem rolls the log forw)
323.2 442.8 R .49(ard ag)-.1 F .49(ainst the database, reapply-)-.05 F
.682(ing an)323.2 454.8 R 3.181(yc)-.15 G .681
(hanges to on-disk pages that were in memory)-3.181 F .14
(at the time of the crash.)323.2 466.8 R .14
(Since the log contains pre- and)5.14 F .957
(post-change state for transactions, the reco)323.2 478.8 R -.15(ve)-.15
G .956(ry system).15 F 1.14(also uses the log to restore an)323.2 490.8
R 3.64(yp)-.15 G 1.14(ages to their original)-3.64 F 1.615(state if the)
323.2 502.8 R 4.115(yw)-.15 G 1.615
(ere modi\214ed by transactions that ne)-4.115 F -.15(ve)-.25 G(r).15 E
(committed.)323.2 514.8 Q 2.051
(If the data disk is lost, the system administrator can)323.2 531 R .887
(restore the most recent cop)323.2 543 R 3.386(yf)-.1 G .886
(rom backup.)-3.386 F .886(The reco)5.886 F(v-)-.15 E 1.298
(ery system will roll the entire log forw)323.2 555 R 1.298(ard ag)-.1 F
1.298(ainst the)-.05 F 2.64
(original database, reapplying all committed changes.)323.2 567 R 4.363
(When it \214nishes, the database will contain e)323.2 579 R -.15(ve)
-.25 G(ry).15 E .535(change made by e)323.2 591 R -.15(ve)-.25 G .534
(ry transaction that e).15 F -.15(ve)-.25 G 3.034(rc).15 G(ommitted.)
-3.034 E .494(If the log disk is lost, then the reco)323.2 607.2 R -.15
(ve)-.15 G .495(ry system can use).15 F 1.853
(the in-memory copies of log entries to roll back an)323.2 619.2 R(y)
-.15 E .026(uncommitted transactions, \215ush all in-memory database)
323.2 631.2 R 1.659(pages to the data disk, and shut do)323.2 643.2 R
1.659(wn gracefully)-.25 F 6.658(.A)-.65 G(t)-6.658 E 2.204
(that point, the system administrator can back up the)323.2 655.2 R .039
(database disk, install a ne)323.2 667.2 R 2.539(wl)-.25 G .039
(og disk, and restart the sys-)-2.539 F(tem.)323.2 679.2 Q EP
%%Page: 6 6
%%BeginPageSetup
BP