"... In PAGE 11: ...able 2- 1 . Differences between replicated and centralized architectures ..................... .... 11 Table4 - 1 .... In PAGE 11: ...able 4- 1 . Protocols for publishing objects into the global space ..................................... 52 Table4 -2 . Protocols for specifying the type of distribution and concurrency control for an object .... In PAGE 11: ...bject ............................................. .... ................................................................. 56 Table4 -3 .... In PAGE 62: ...a dictionary which exists on each site. The operations for manipulating the ObjectDirectory is indicated by their method protocols as given in Table4 -1. For readers not familiar with Smalltalk a method is an operation on the object that owns it.... In PAGE 66: ...bject can be made sharable (e.g. Smalltalk apos;s current implementations of a Dictionary, OrderedCollection or Rectangle). The protocols used to specify the distribution strategy (#replicated #replicatedLocking or #centraIized shown in Table4 -2) are applied to an instance of an object afier it has been created. For example, we can create a replicated Rectangle by sending an instance of a rectangle object the message #replicated.... In PAGE 67: ... When the replicating object is referenced remotely, a copy is sent I that is automatically kept up to date. Table4 -2. Protocols for specifying the type of distribution and concurrency control for an object.... In PAGE 68: ... The developer needs to be able to specify the message, the type of event that it creates, whether the event should be generated before or after the message is executed, and what call-backs are attached to the event. The protocol #adiiPreGroupwareEventronMessage:, shown in Table4 -3, causes the receiving object to generate the specified event (aSymbol) before the message with the specified name (aMessageName) is executed. Similarly, #addPostGroup wareEvent:onMessage: causes the event to be generated @er the message has executed.... In PAGE 69: ... Specifies that the call-back associated with anobject and selector uSelector be removed from the event List. Table4 -3. Protocols for adding events to shared objects.... In PAGE 71: ... The API. Table4 -4 shows the protocols for environments which we call GroupEnvironments. The hew message lets the developer create an environment.... In PAGE 71: ... If the environment does not contain astring, search the super environment. - - - - - - - - - - - - - - Table4 -4. Protocols for using environments in groupware applications.... ..."

"... In PAGE 12: ... Local political conditions in many developing countries have led to extensive campaigns by local terrorist groups. Table1 provides a summary of some terrorist organizations operating in developing countries, based on data from the US ... ..."

"... In PAGE 3: ... Only in figures for the private sector, which double those for the public sector, does Catalonia approach averages for Eu- rope. Table4 shows all public expenditure directed to chem- istry in Catalonia during the considered period. The table in- cludes the research funds allocated by European, Spanish and Catalan agencies to all institutions engaged in research activi- ties in chemistry.... ..."

"... In PAGE 10: ...However, as Table1 illustrates, that the global distribution of these resources is far from even. Figures on financial flows among civil society organizations are hard to come by, but the information available paints a fairly clear picture.... ..."

"... In PAGE 7: ... global pointers are references to a global variable and an alias of the global within the same scope. Table5 organizes the incidental aliasing according to these... ..."

"... In PAGE 14: ... The joint experimental infrastructure we envision is shown in Figure 4. We have already started to work with various groups, shown in Table2 , to integrate existing infrastructure into this joint facility. Encouraged by our experimental results at SC2002, we started planning for the next experiment in the middle of the Conference back in November 2002.... ..."

"... In PAGE 13: ...Results from this parameterization for sulfate cloud forcing using all carbonaceous aerosols as pre-existing aerosols are shown in Table3 .... In PAGE 13: ... These particles simulate a possible source of marine background particles such as sea salt or marine sources of organic aerosol. The global average annual mean radiative forcing is listed in Table3 . The cloud forcing by carbonaceous aerosols decreased from - 4.... ..."

"... In PAGE 13: ... Distinguishing at least among sub-orders improves regional estimates of regional variation in organic- C stocks. Other published means for Mollisols or for tem- perate grassland soils that have been applied to regional and global extrapolations range from 9 to 19 kg C m-2 ( Table5 ). The low estimate results from including only the top 50 cm of soil (Kimble et al.... ..."

"... In PAGE 80: ...hich is to support the Fa.m.i.1t.y System, several design techniques can be employed; These can be categorized as follows: File Organization Choices: pile, sequential, indexed-sequential, indexed, direct or ring; File Partitioning Choices: One record per tuple, or one record per segment; Use 01 apos; auxiliary access files; In practice all three choices will be employed; In order to establish a baseline; the nerf apos;ormance of a pile of unpartitioned records without auxiliary access files will be presented initially; Table4 -1 summarizes the basic relations and their parameters, as derived from the data presented in Table 3-1; Since all files are evaluated independently, the size of the ruling part has been added to the recordsize; The primary files summarized in Table 4-1 do not include the lexicons and referenced entity files; The space required for these is presented in Table 4-3 using the assumptions that they ... In PAGE 80: ...hich is to support the Fa.m.i.1t.y System, several design techniques can be employed; These can be categorized as follows: File Organization Choices: pile, sequential, indexed-sequential, indexed, direct or ring; File Partitioning Choices: One record per tuple, or one record per segment; Use 01 apos; auxiliary access files; In practice all three choices will be employed; In order to establish a baseline; the nerf apos;ormance of a pile of unpartitioned records without auxiliary access files will be presented initially; Table 4-1 summarizes the basic relations and their parameters, as derived from the data presented in Table 3-1; Since all files are evaluated independently, the size of the ruling part has been added to the recordsize; The primary files summarized in Table4 -1 do not include the lexicons and referenced entity files; The space required for these is presented in Table 4-3 using the assumptions that they ... In PAGE 80: ...hich is to support the Fa.m.i.1t.y System, several design techniques can be employed; These can be categorized as follows: File Organization Choices: pile, sequential, indexed-sequential, indexed, direct or ring; File Partitioning Choices: One record per tuple, or one record per segment; Use 01 apos; auxiliary access files; In practice all three choices will be employed; In order to establish a baseline; the nerf apos;ormance of a pile of unpartitioned records without auxiliary access files will be presented initially; Table 4-1 summarizes the basic relations and their parameters, as derived from the data presented in Table 3-1; Since all files are evaluated independently, the size of the ruling part has been added to the recordsize; The primary files summarized in Table 4-1 do not include the lexicons and referenced entity files; The space required for these is presented in Table4 -3 using the assumptions that they ... In PAGE 83: ...application programs, workspaces for sorting and reorganization of files, etc;; At this point 2,000,000 bytes will be allocated to this function; INITIAL PERFORMANCE DESIGN CONSIDERATIOIiS From Table4 -1 and 4-2 the following observations can be made: The aggregate file size is such that simple sequential file organization methods use an excessive amount of storage space; Much space in the sequential file organization is due to the maximum length of the notes (800 characters). The files in either case are of such a size that high capacity file storage devices will be necessary.... In PAGE 87: ...seen during the AAMRS study [2]; Pharmacy Labels Automatic Referral Letters Statistical Analysis of Data Graphical Presentation of Data The usage due to the services specified can be a~plied to the Family System files as shown in Table4 -6; This table is obtained by transposition of the data from Table 4-5; R indicates file read U indicates file read and updated X indicates exhaustive file reading Y indicates reorganization; The monthly report generation, since it can be distributed over the weekends, has been taken as a weekly load of 12/52=0;23 of the ... In PAGE 87: ...seen during the AAMRS study [2]; Pharmacy Labels Automatic Referral Letters Statistical Analysis of Data Graphical Presentation of Data The usage due to the services specified can be a~plied to the Family System files as shown in Table 4-6; This table is obtained by transposition of the data from Table4 -5; R indicates file read U indicates file read and updated X indicates exhaustive file reading Y indicates reorganization; The monthly report generation, since it can be distributed over the weekends, has been taken as a weekly load of 12/52=0;23 of the ... In PAGE 93: ... . . .. . .. -- Grand totals 39698 (10;2) 11016 (98;07) (9) w-7 -- - A review of Table4 -6 can verify the data flow in the Family System files; The verification shows that all files are used and updated; Most files are updated only during one function; In order to control the data flow where files are updated by more than one function and avoid deadlock [I :Ch;13;2], the control over data will have to be respecified in finer units; Two choices are possible : 1) control by record segmenting 2) control by time constraints For instance, the indicators that a patient is to be placed on a flowsheet protocol, as given above as 300 Patient d/U = 20 reside in a segment of the Patient file which is not to be updated by any other function; On the other hand the updating of flowsheet entries after a visit, 400 Visits c/U = 50 is delayed in time until all other encounter activity has been ... In PAGE 94: ...Update(U), and Reorganize(Y) activity were recorded Table4 -6; it should be noted that the totals shown are not particularly useful for performance evaluation; For each file the time used for each action will depend on the file organization, its size, and the recordsize to be retrieved; In particular, Read-entire(X) and Reorganize(Y) are dependent on file size; From the grand total of Read(R) and Update(U), namely approximately 50,000 actions/day; it is evident, however, that each request has to be fulfilled in considerably less than a second; since there are only 28,800 seconds in a working day; In the next section of this chapter the file design will be refined with this objective in mind; STORAGE OF NOMINAL DATA Omitting the notes or removing them to indirect storage, as paper or microform, could reduce storage requirements considerably; Notes appear in the following files: - ,~abie h- apos;; Stored Notes File Name n Notes (source) Number . - 30 0 Patients 30000 2(3,4) 4:3 Other Allergy 6000 14-1 5; ;I Actlon 270000 1[5:7:1) 888 ibiedication Bon-drug Therapy 2 Tests Ordered I; Problem Notes Total Notes ... In PAGE 96: ...independent of the manner in which the notes are kept, the notes within the files will be replaced with reference pointers; These could then be used to refer to notes stored on a cheap high volume file, to a microform storage, or to a payer document; Only in the Other Allergy File (4~3). which is relatively small, will the notes be retained since rapid reference here may be essential; With further experience a suitable encoding may also reduce the average note size; The files affected by keeping notes separate are listed in Table4 -8; These entries replace the equivalent entries of Table 4-1; Table 4-8 Files Shrunk by Removing Notes Name Size rd For a n ~FFRf av/WvIavg ,max) 30 0 Patients 3000(2) 13/29 $4 ; 1 Actions 270000 9/2 1 Medications 300000 1 /20 - 900 Non-drug Therapy 1 0000 2/14 - 1112 Tests Ordered 2?0000 11/25 - I I ; 13 Problem Note 180000 5/ 13 - Storage reduction 1,146,000 notes or (pile file org;) 87,096,000 characters (about 2 disk units) (seq; file org;) ... In PAGE 96: ...independent of the manner in which the notes are kept, the notes within the files will be replaced with reference pointers; These could then be used to refer to notes stored on a cheap high volume file, to a microform storage, or to a payer document; Only in the Other Allergy File (4~3). which is relatively small, will the notes be retained since rapid reference here may be essential; With further experience a suitable encoding may also reduce the average note size; The files affected by keeping notes separate are listed in Table 4-8; These entries replace the equivalent entries of Table4 -1; Table 4-8 Files Shrunk by Removing Notes Name Size rd For a n ~FFRf av/WvIavg ,max) 30 0 Patients 3000(2) 13/29 $4 ; 1 Actions 270000 9/2 1 Medications 300000 1 /20 - 900 Non-drug Therapy 1 0000 2/14 - 1112 Tests Ordered 2?0000 11/25 - I I ; 13 Problem Note 180000 5/ 13 - Storage reduction 1,146,000 notes or (pile file org;) 87,096,000 characters (about 2 disk units) (seq; file org;) ... In PAGE 98: ...method; the wasted mace per record is then and the bulk transfer rate is The records specified for the Family System [3] have a well-defined variable content; Because of this fact records will not contain attribute names in any of the files being considered, althou5h the MUMPS files will include the descriptive subscript values for the globals [ 1 : Chi4 ;5;2] ; For variable sized records, the length (Rvar) includes field separation markers; The number of weekly additions (0) of records to the files is based on the data from the Family System [3:Ch;2], as given in Tables 3-1 and 3-3; Subsequent evaluations in this chapter find that the parameters T (get-next data record) and T N u (update a field of a data record) are of minor importance due to the usage patterns expected; Search for the next record is an important activity in subset searching and is associated with complex data analysis; Medical records are mainly maintained by adding to the files so that updating of old data records is infrequent; Descriptive File Parameters: For the various primary files to be considered, the descriptive parameters, when computed using the conditions cited above, are displayed in Table4... In PAGE 100: ...closely the performance of a data base system which is built on apos;relational principles apos; [1:Ch;9;11 and which has not been further augmented with ancilliary access paths; The results for the Primary Files are shown in Table4 -10; Table 4-10 Pile File Performance Measures File No; T ;T T F N I The retrieval times for the primary files using the pile file organization are for most files much greater than allowable as established by the approximate criterion derived from Table 4-6; ... In PAGE 100: ...closely the performance of a data base system which is built on apos;relational principles apos; [1:Ch;9;11 and which has not been further augmented with ancilliary access paths; The results for the Primary Files are shown in Table 4-10; Table4 -10 Pile File Performance Measures File No; T ;T T F N I The retrieval times for the primary files using the pile file organization are for most files much greater than allowable as established by the approximate criterion derived from Table 4-6; ... In PAGE 100: ...closely the performance of a data base system which is built on apos;relational principles apos; [1:Ch;9;11 and which has not been further augmented with ancilliary access paths; The results for the Primary Files are shown in Table 4-10; Table 4-10 Pile File Performance Measures File No; T ;T T F N I The retrieval times for the primary files using the pile file organization are for most files much greater than allowable as established by the approximate criterion derived from Table4 -6; ... In PAGE 101: ...T = gt;T L Total FP HP P P = 75009 + 357 = 75,366 seconds or 20;94 hours = 20 hours 56 minutes per day It can be seen, however, that files ;!$ Entitlements Drug Allergies 1 5 Appointments Da Sheet 1 apos;3 ; I ~azients per Office are such that no further file organization can improve their performance significantly; The minimal read access time for the hardware be in^ considered is s + r + B/t = ;058 sec; and the minimal update access time is s + 3r + B/t = ;092 sec; They will be omitted from further consideration and can either be implemented as simple pile files or according to any other convenient general organization; As pile files their aggregate time requirements will be as shown in Table4 -11; Values used are T or T as appropriate for the services being performed; I ... In PAGE 102: ... Table4 -11 Time to be Allocated to Small Files T L TLL T L T L F R X Xd Xw I/U U YY Entl ;I17 950 ;355 1 1 ;092 20 na - DrAl ;2 7 450 ;4 4 - 1 ;092 na - ~ppt :0 amp;2 370 :164 I 4:23 :208 38 2327 I ;o 2 - I ::el8 :0?2 %go 1% 1 I $8 88 12 - daily time = T L + T L + T L F R X Xd I/U U = 344 + 1 + 190 = 533 sec = 9 min/day weekly time = T L + T L = 2 sec/veek xxw YY Sequential, Indexed-Sequential, and Direct File Organizations: Faster access according to one attribute is afforded by the sequential; indexed-sequential, and direct file organizations; Of major interest are here the retrieval speed T and the file update F speed T ; Individual record update is of interest only for the I Appointment Detail file (12;l); The values obtained for the pile file organization in regard to T and T can provide initial X Y ... In PAGE 103: ...For access to the sequential file [1:Ch;3;2;31, the algorithm will consist of a binary search through the main file and a serial search through the transaction iog i apos;ile; NO ~ur apos;f apos;er is kept available for the transaction file so that the update response time is equal to that for a pile file; All updates are collected into transection files for the day, and at the end of the day these files will be sorted and merged into the main sequential files; The number of daily updates is based on the daily load L u and is related to the file growth, o; For the primary files the time required for this process. T , will be mainly a function of C the file sizes; The sort phases will require approximately T = L log L (T +T sort U 2U F I where T and T pertain to small pile files; F I The merge phases will require approximately T = (n+L ) ~/t apos; and T =T =T +T merge U C Y sort merge The value of L can be obtained from Table4 -6; An average bulk U transfer rate, t apos;= 220,000 bytes/second; gives T =;052, T =;092 F I files which contain the daily update (0/5) transaction; ... In PAGE 104: ...Table4 -12. T r log (n R/U) 1 (s+r+B/t) + 1/2 0/5 ~/t apos; F 2 Indexed-Sequential: An indexed-sequential file organization provides faster access through a tree search of key values [I:Ch;3;2;31; Its efficiency depends on a small number of updates per file; Some additional space, SI; is required; This space is a function of the fanout ratio, y, [I:Ch;3;3;1], blocking; and number of records; n, for a two-level (x=2) index; File 12;l only requires one level; The values for Rr depend on the size of the ruling part and are given in Table 3-7; T =s + (2 + Pov (1+1/2~ov)(r+B/t) ; Pov = o/(n+o) F T = ( R/B + 2 Pov (1-R/B) )(r+B/t) (two buffers are available) ... In PAGE 105: ...Table4 -12 Sequential and Indexed-Sequential File Performance Measures I----- Sequential -------I I ---- Indexed-Sequential ----I File No; Total time for daily cleanup 3620;O Total space for index 494,100 or approx; 1 hour Direct File Organization: For a direct file the performance depends mainly on the extra storage allocation (m/n) and on the bucketsize [I:Ch;3;5;3]; With a fixed blocksize the blocking factor B/R is especially important; With unspanned blocking of records there is an additional loss of space so that required space is recalculated here as 11 basic filesize = (- ) B r B/H 1 T is not available; N p = funct(B/R,m/n) ZLS shown in Figure 2-2; Larger values of m/n increase the storage cost and this increase ... In PAGE 106: ...Before N evaluating the direct file, the remaining primary files will be reviewed in order to eliminate from analysis of direct files those for which serial access is important; The services to be considered were listed earlier in Table4 -5; Table 4-13 Suitability for Direct Access Service . .... In PAGE 106: ...Before N evaluating the direct file, the remaining primary files will be reviewed in order to eliminate from analysis of direct files those for which serial access is important; The services to be considered were listed earlier in Table 4-5; Table4 -13 Suitability for Direct Access Service . .... In PAGE 106: ...e B1 1 Preparation p) System ~ana~ement System Maintenance Plessaaes Keyed on individual patient name; 11 11 It 11 11 Serial on day sheet (1311); Keyed $n individual pattent; II II II It record; 11 I1 I1 tt and test; Serral a cess on ppointment flies F12, 1238 Wegkly se~ial acgess gn vigits (420); 11 11 11 problems seen (11;l); Serial access on a pointment (12;l) Keyed on indivi.dua? patient; Read entire Patient file; Read file 3b serially; Read file 12 and non-primary files serially; Read non-primar files; Read files 17; r8 serially A review of these requirements show that of the primary files only files Visits (400) , Problems Seen ( 11 ; 1 ) , and Appointment Detail (12;l) should be eliminated as candidates for direct file organization; The remainina primary files are presented in Table4... In PAGE 107: ...Table4 -14 Direct File Performance Measures I 1~05 ---- I 1;10 ----I 1;25 ----I I ---- 1 ---- I ---- T I File B/R Basic T T T T T No; Filesize F I F I F I TotalFile 67,139,961 7046 60 Increase - 3:3863899 7 85 ,95 ?:711,992 B2:36t:85: Summary of Sequential, Indexed-Sequential, and Direct Primary Files: A review of Table 4-12 shows that the sequential file organization with a binary search still does not reduce the access time for most files to a comfortable limit ( lt; lt; 1 second); The additional hour required to clean up the files at the end of every day is another liability; although this operation could be combined with the creation of back-up files; Given a reasonably reliable operation and adequate performance of the Transaction Log file such an effort would not be required every day for backup purposes alone; An indexed-sequential file, although more complex, shows a more acceptable level of performance; The penalty of space for the ... In PAGE 107: ...Direct File Performance Measures I 1~05 ---- I 1;10 ----I 1;25 ----I I ---- 1 ---- I ---- T I File B/R Basic T T T T T No; Filesize F I F I F I TotalFile 67,139,961 7046 60 Increase - 3:3863899 7 85 ,95 ?:711,992 B2:36t:85: Summary of Sequential, Indexed-Sequential, and Direct Primary Files: A review of Table4 -12 shows that the sequential file organization with a binary search still does not reduce the access time for most files to a comfortable limit ( lt; lt; 1 second); The additional hour required to clean up the files at the end of every day is another liability; although this operation could be combined with the creation of back-up files; Given a reasonably reliable operation and adequate performance of the Transaction Log file such an effort would not be required every day for backup purposes alone; An indexed-sequential file, although more complex, shows a more acceptable level of performance; The penalty of space for the ... In PAGE 108: ...Table4 -14 presents the performance of direct files; For many of the files listed the performance is yet better, even at very low (1;05) excess spacc ratios; Only for files with large records (B/R lt; 10) is the performance less than that for indexed-sequential files; The extra space requirements are, however; in any case greater than those required for indexes; In order to comnare the three file organizations, the total daily usage time, UT, will be calculated; for t=S zential) IS [ %exed-sequential) D (direct) and p identifies the various files; The following considerations apply to this summary calculation: For the sequential file, L =I for all files due to the cleanup required ; and x P T =T for these files XSP XP For the direct files, the files identified with p = 300 uses m/n = 1;10 and p = 4;3 uses m/n = 1;25 All other files use m/n = 1;05 Also T = m/n T where applicable to account for the extra XDP CP space to be read in a direct file exhaustive read operation ... In PAGE 109: ...for those files where a direct organization was not found feasible according to Table4 -13, so that T =T 9 FDp FISp T =T , and XDp XISp T =T 9 IDp IISp for files identified with p = 400; 11;1, 12 With these considerations the daily primary file usage for the three file organizations is UT = 14515 + 3620 + 265 = 18400 sec/day or 5 hours 6 mins/day S UT = 837 + 392 + 504 = 1733 sec or 29 midday IS UT = 696 + 429 + 389 = 1541 sec or 25 midday D This shows even more clearly that a sequential file, even with binary search and deferred update; is not adequate; The load presented by indexed-sequential or direct file for thjs aspect of file services is still significant; Another nine minutes per day will be required to serve the small primary files of Table 4-11, and yet more time is needed to provide for usage of the Lexicons; Referenced Entity Files; and Service Files; The indexed-sequential and direct file organizations show a space versus speed tradeoff; F apos;or the 12 files of Table 4-14, at the m/n ratios chosen, the additional space required for direct files is m- n SAD =z - filesize = 5;306;998 bytes; n DP ... In PAGE 109: ...for those files where a direct organization was not found feasible according to Table 4-13, so that T =T 9 FDp FISp T =T , and XDp XISp T =T 9 IDp IISp for files identified with p = 400; 11;1, 12 With these considerations the daily primary file usage for the three file organizations is UT = 14515 + 3620 + 265 = 18400 sec/day or 5 hours 6 mins/day S UT = 837 + 392 + 504 = 1733 sec or 29 midday IS UT = 696 + 429 + 389 = 1541 sec or 25 midday D This shows even more clearly that a sequential file, even with binary search and deferred update; is not adequate; The load presented by indexed-sequential or direct file for thjs aspect of file services is still significant; Another nine minutes per day will be required to serve the small primary files of Table4 -11, and yet more time is needed to provide for usage of the Lexicons; Referenced Entity Files; and Service Files; The indexed-sequential and direct file organizations show a space versus speed tradeoff; F apos;or the 12 files of Table 4-14, at the m/n ratios chosen, the additional space required for direct files is m- n SAD =z - filesize = 5;306;998 bytes; n DP ... In PAGE 109: ...for those files where a direct organization was not found feasible according to Table 4-13, so that T =T 9 FDp FISp T =T , and XDp XISp T =T 9 IDp IISp for files identified with p = 400; 11;1, 12 With these considerations the daily primary file usage for the three file organizations is UT = 14515 + 3620 + 265 = 18400 sec/day or 5 hours 6 mins/day S UT = 837 + 392 + 504 = 1733 sec or 29 midday IS UT = 696 + 429 + 389 = 1541 sec or 25 midday D This shows even more clearly that a sequential file, even with binary search and deferred update; is not adequate; The load presented by indexed-sequential or direct file for thjs aspect of file services is still significant; Another nine minutes per day will be required to serve the small primary files of Table 4-11, and yet more time is needed to provide for usage of the Lexicons; Referenced Entity Files; and Service Files; The indexed-sequential and direct file organizations show a space versus speed tradeoff; F apos;or the 12 files of Table4 -14, at the m/n ratios chosen, the additional space required for direct files is m- n SAD =z - filesize = 5;306;998 bytes; n DP ... In PAGE 110: ...SIS = SI P P = 436;000 bytes or 4 ; 87 1 ; 000 bytes less; The direct file organization applied to these 12 files saves 4 min/day at $;20/minute cost for the entire system; This equals $360/year and increases storage cost by 4;871;000 bytes at $300/Mbyte year or $1461;30/year; With these values the indexed-sequential file seems to be preferable; but not to an overriding extent; When a specific file can make a significant contribution to system performance; then a direct file organization may well be justified; RING OR HIERARCHICAL FILES The sequential, the indexed-sequential, and the direct file depend on fixed size records, and there is hence an additional space penalty over the pile file; This penalty was documented following Table4 -8 and amounted for all primary files to 117,129;000 bytes versus 56,955,000 bytes; It remains hence desirable to investigate file organization methods which handle variable length data more efficiently; Prime candidates are here the indexed file organization [1:Ch;3;4;31, the ring file organization [ 1:Ch;3;4;63, and the MUMPS file structure [ I:Ch;4;5;23; Of special interest here are the Family System files that have a natural hierarchical structure, as shown by the data model presented as Figure 3-4; Table 4-15 presents again the familiar primary files; now with the parameters which are important to ring ... In PAGE 110: ...SIS = SI P P = 436;000 bytes or 4 ; 87 1 ; 000 bytes less; The direct file organization applied to these 12 files saves 4 min/day at $;20/minute cost for the entire system; This equals $360/year and increases storage cost by 4;871;000 bytes at $300/Mbyte year or $1461;30/year; With these values the indexed-sequential file seems to be preferable; but not to an overriding extent; When a specific file can make a significant contribution to system performance; then a direct file organization may well be justified; RING OR HIERARCHICAL FILES The sequential, the indexed-sequential, and the direct file depend on fixed size records, and there is hence an additional space penalty over the pile file; This penalty was documented following Table 4-8 and amounted for all primary files to 117,129;000 bytes versus 56,955,000 bytes; It remains hence desirable to investigate file organization methods which handle variable length data more efficiently; Prime candidates are here the indexed file organization [1:Ch;3;4;31, the ring file organization [ 1:Ch;3;4;63, and the MUMPS file structure [ I:Ch;4;5;23; Of special interest here are the Family System files that have a natural hierarchical structure, as shown by the data model presented as Figure 3-4; Table4 -15 presents again the familiar primary files; now with the parameters which are important to ring ... In PAGE 111: ...Table4 -15 Hierarchical Parameters Filesize Recordsize Fanout File Filename ParentEntryLevel n Ravg a y The average natural fanout, y; appears to be very low; Block-Oriented Hierarchies: In some file implementations the rings or hierarchies are defined by the use of block pointers [1:Ch;2;3;31; Each subsidiary ring will. occupy a single block chain in a system which does not allow sharing of blocks; and at least one block is ... In PAGE 112: ...ring may require many blocks; The total number of blocks is then I level lt; x ; and y gt; 1 P P I level lt; x , and y lt; 1 P P = 833,724 blocks This is equivalent to 1667 i4bytes of file storage or 34 disk units for the primary files alone, so it is obvious that the natural hierarchy will have to be reshaped if blocks cannot be shared; A MUMPS Implementation of the Family System: A mCIFlPS system i 1 :Ch;4;5;21 is an example of such a block-oriented hierarchy; While the storage requirements will be excessive, the same data base structure used up to this point will be used to illustrate the corresnonding structure using MUWS globals; The structure of MUMPS also forces the use of a distinct level for ruling parts unless denesting hs taken place; The transformation of the data base model to a MUMPS tree is shown in Table4 -16; The level numbers are assigned beginning from the directory Level; d; Each level of the MUMPS hierarchy is related to the number of subscripts in the global variable; Each global variable defines ... In PAGE 117: ...The blocksize of MUMPS systems is less than the blocksize assumed uu to this point, typically 512 or 768 bytes; But even these small blocks will be rarely well utilized if the original data base model is used; Figure 4-1 illustrates the layout for part of the primary hierarchy; In a file structure with linkages the space for the linkage pointers has to be accounted for; The linkage pointers refer to subsidi3.r~ se~ments and the number of pointers at a level can hence be obtained as a; ptr 2 level apos;level- I In order to locate the individual fields of the qlobals, PIUMPS uses a two-byte identification; Where the ruling part of the tuple contains only an identification field, no distinct value field will be required; The two requirements are combined in Table4 -17 in the column pid where H = 2a + 2a pid ptr ... In PAGE 119: ...ikr. blocks per segment bs=r~ /~l seg total blocks b =nbs The values in the Table4 -17 are evaluated for B = 500; One of the goals of the vairable-length segment architecture, better utilization of disk space is hence not achieved by a direct implementation of the data base model; The low density is evidenced by the fact that R lt; lt; B for most of the segments shown seF1; in Table 4- 17; The value of R for the MUblPS file organization did not affect pid the actual total file size significantly; The increase of segment size, as the file is now constituted, with a very low average block density will cause only very few blocks to overflow; The values of the subscript fields used to describe the entries also replaces some of the coded data fields now used to identify ... In PAGE 119: ...ikr. blocks per segment bs=r~ /~l seg total blocks b =nbs The values in the Table 4-17 are evaluated for B = 500; One of the goals of the vairable-length segment architecture, better utilization of disk space is hence not achieved by a direct implementation of the data base model; The low density is evidenced by the fact that R lt; lt; B for most of the segments shown seF1; in Table4 - 17; The value of R for the MUblPS file organization did not affect pid the actual total file size significantly; The increase of segment size, as the file is now constituted, with a very low average block density will cause only very few blocks to overflow; The values of the subscript fields used to describe the entries also replaces some of the coded data fields now used to identify ... In PAGE 120: ...Table4 -17 MUMPS Storage Requirements a R File, part n ------------------- ----------------- blocks ptrs values fields avg pid seq; . .... In PAGE 122: ...low level ruling parts unnecessary, since these connections are explicitly expressed by pointers; Omitting these attributes, however, has an undesirable effect on structural integrity, since now the linkages become essential [1:Ch;13;3;2]; If the hierarchy is transformed to provide a higher storage density, then the space taken by descriptors and pointers will have to be reconsidered; The access speed in MUMPS system is mainly a function of the number of levels; Given full compute and file accessing overlap, and an optimal record placement (possible when the files are not densely utilized), the access speed can be as fast as T us + r + r1/2 bsl (B+w)/~ apos; for the top level Fd and T = T + r1/2 bsl (B+w)/~ apos; for level i = d to bottom-1 Fi-1 Fi Note that the use of t apos; is appropriate since the search sequence extends contiguously over block boundaries; Table4 -18 Levels in a MUMPS File Tree Level Dependent Parts bs ... In PAGE 123: ...A reconstitution of the data base model in order to reduce the excessive storage requirements will increase the expected fetch times; An optimal design from the performance point of view may achieve an acceptable balance of access speed and filespace; the data base model will then be quite different and should be re-evaluated to assure that it satisfies the needs of the applications which led to its construction; Record-Oriented Hierarchies: In a ring structures file which allows multiple record types per block; the storage can be used very efficiently; Only space for pointers, record descriptors, and record separators is needed, and as pointed out earlier, the pointers can replace some redundant ruling part attributes if essential links are ~ermissible; The storage requirement will then be very similar to that of the pile file ( Table4 -2); The access problem to the top level (x) Is solved in ring-structurled file system generally by the use of direct access; The results from Table 4-14 are knce usable for files 1, 12, 13; Subsidiary rings are accessed by following individual pointers; ... In PAGE 123: ...A reconstitution of the data base model in order to reduce the excessive storage requirements will increase the expected fetch times; An optimal design from the performance point of view may achieve an acceptable balance of access speed and filespace; the data base model will then be quite different and should be re-evaluated to assure that it satisfies the needs of the applications which led to its construction; Record-Oriented Hierarchies: In a ring structures file which allows multiple record types per block; the storage can be used very efficiently; Only space for pointers, record descriptors, and record separators is needed, and as pointed out earlier, the pointers can replace some redundant ruling part attributes if essential links are ~ermissible; The storage requirement will then be very similar to that of the pile file (Table 4-2); The access problem to the top level (x) Is solved in ring-structurled file system generally by the use of direct access; The results from Table4 -14 are knce usable for files 1, 12, 13; Subsidiary rings are accessed by following individual pointers; ... In PAGE 125: ...With these estimates the performance of a ring-structured file organization can be presented;: Table4 -19 Ring Structure Performance File Access direct via 1 direct - via 12 direct via 13~1 5;1 0% from Table 4-12 19?2;7135 4 64- 072 6a9 1 8457 na ;1722+ from Table 4-11 339;7305 -10 2~ from Table 4-11 10 ~7983 3 Direct files are taken with m/n = 1;05; The total operation of the ring-structured file requires then = 20044;Ci sec/day = 5;57 hours/day ... In PAGE 125: ...With these estimates the performance of a ring-structured file organization can be presented;: Table 4-19 Ring Structure Performance File Access direct via 1 direct - via 12 direct via 13~1 5;1 0% from Table4 -12 19?2;7135 4 64- 072 6a9 1 8457 na ;1722+ from Table 4-11 339;7305 -10 2~ from Table 4-11 10 ~7983 3 Direct files are taken with m/n = 1;05; The total operation of the ring-structured file requires then = 20044;Ci sec/day = 5;57 hours/day ... In PAGE 125: ...With these estimates the performance of a ring-structured file organization can be presented;: Table 4-19 Ring Structure Performance File Access direct via 1 direct - via 12 direct via 13~1 5;1 0% from Table 4-12 19?2;7135 4 64- 072 6a9 1 8457 na ;1722+ from Table4 -11 339;7305 -10 2~ from Table 4-11 10 ~7983 3 Direct files are taken with m/n = 1;05; The total operation of the ring-structured file requires then = 20044;Ci sec/day = 5;57 hours/day ... In PAGE 125: ...With these estimates the performance of a ring-structured file organization can be presented;: Table 4-19 Ring Structure Performance File Access direct via 1 direct - via 12 direct via 13~1 5;1 0% from Table 4-12 19?2;7135 4 64- 072 6a9 1 8457 na ;1722+ from Table 4-11 339;7305 -10 2~ from Table4 -11 10 ~7983 3 Direct files are taken with m/n = 1;05; The total operation of the ring-structured file requires then = 20044;Ci sec/day = 5;57 hours/day ... In PAGE 127: ...Index (19); This file will be used as an example. Detailed Design of a Direct File: The parameters which control the desipn of a direct file are [1:Ch;3;51; the number of records, n the average recordsize, R the space available for records; mR the size of the buckets, B The known parameters have the values The request rate for the Patient Name Index is found from Table4 -6 as L =700; From [1 :Ch;3;4;31 it can be seen that the R ~ime required per access is 1 n T =(I+-- )(s+r+B/t) for B = R F 2 m-n or T = (l+p)(s+r+9/t) with p = t apos;unct(m/n, B/R) F The equivalent cost is: T L cost R R system The disk storage is: m (R+W) cost where W = G R/B [1:Ch;2;2;4] disk The core storage cost for large blocks is: T L (B-R) cost F R ... In PAGE 128: ...show these costs for a number of values of B/R and m/n and typical val.ues of s (;035 sec), r (;017 sec), and a high speed disk t(620000/sec); cost = $0;20/min system cost = $300/Mbyte year disk cost = $200000/Mbyte year core G = 193 bytes c = 1 second Table4 -18a Fetch Cost Table 4-18b ... In PAGE 128: ...show these costs for a number of values of B/R and m/n and typical val.ues of s (;035 sec), r (;017 sec), and a high speed disk t(620000/sec); cost = $0;20/min system cost = $300/Mbyte year disk cost = $200000/Mbyte year core G = 193 bytes c = 1 second Table 4-18a Fetch Cost Table4 -18b ... In PAGE 129: ...Table4 -18c Incremental Core Memory cost/day = ( l+(s+r+B/t)*p) (B-R) 700*;000667/8/60/60 n/n B/R B-H . .... In PAGE 129: ..... Table4 -18d Total Cost/Day ... In PAGE 130: ...a) The fetch cost decreases with increasing m/n and increases slightly with increasing blocksize; When m/n gt; 1;1 and B/R gt; 10 the effects become minimal; b) The storage cost is significant; The minimum is reached at high B/R and low m/n; The effect of increasing blocksize diminishes as B/R gt; 50; c) The core memory cost for the small amount of time that the buffer is active is low; It is minimally affected by increased m/n for m/n gt;= 1;05; d) If slower devices are used, then the loss in access speed can only be slightly compensated by increased storage since the selected optimum is already close to the minimum ( Table4 -18a), and a different blocking factor has no effect; The minimum total cost is found at B/R = 100 or blocksize b = 4900 bytes m/n = 1;05 or storage = 30000 * 1;05 * 49 = 1;54 Mbytes This is a fairly large blocksize, and it may be desirable to use a smaller blocksize in order to simplify memory management; The cost of a smaller blocksize, as shown in the column for m/n = 1;05 in Table 4-18d does not increase rapidly until B/R lt; 10 or B lt; 490 characters; These results can be extrapolated easily to the other direct files; The usage of the Terms file (900) is very high; ... In PAGE 130: ...a) The fetch cost decreases with increasing m/n and increases slightly with increasing blocksize; When m/n gt; 1;1 and B/R gt; 10 the effects become minimal; b) The storage cost is significant; The minimum is reached at high B/R and low m/n; The effect of increasing blocksize diminishes as B/R gt; 50; c) The core memory cost for the small amount of time that the buffer is active is low; It is minimally affected by increased m/n for m/n gt;= 1;05; d) If slower devices are used, then the loss in access speed can only be slightly compensated by increased storage since the selected optimum is already close to the minimum (Table 4-18a), and a different blocking factor has no effect; The minimum total cost is found at B/R = 100 or blocksize b = 4900 bytes m/n = 1;05 or storage = 30000 * 1;05 * 49 = 1;54 Mbytes This is a fairly large blocksize, and it may be desirable to use a smaller blocksize in order to simplify memory management; The cost of a smaller blocksize, as shown in the column for m/n = 1;05 in Table4 -18d does not increase rapidly until B/R lt; 10 or B lt; 490 characters; These results can be extrapolated easily to the other direct files; The usage of the Terms file (900) is very high; ... In PAGE 131: ...since as noted in paragraph (d) above the fetch minimum is close to the optimum found; At the optimum (m/n=1;05), the overflow probability p= 0;06 [Figure 2-31 ner request leads to The corresponding update time is If these values apply to all Lexicons and Referenced Entity files, then the daily times will be Tday = 1532 sec; ; Tday = 128 sec; F I Files 19, 20, and 21 have to be read in their entirety once daily; Given their basic sizes from Table4 -3, m/n=1;05 and t apos;=220000/sec;, this will require ( 1470000+6000C+50000) m/n Tday = = 8 seconds; X I. apos; so that if these files are optimally configured require Tday = Tday + Tday + Tday = 1668 sec; or 28 min; daily F I X ... In PAGE 132: ...The demand on most of the service f apos;iles is modest; The greatest demand documented in Table4 -6 is on the Transaction Log; If a separate channel is available; then its activity can be largely overlapped [1:Ch;5;4]; Otherwise the parameters for sequential writing will apply; A high security Transaction Log may not buffer outnut; Then every transaction requires (s+r+R/t); The aggregate daily load is then Tday = 850(s+r+28/t)+4450(s+r+ll/t) = 276 sec; = 4 min; 36 sec; SUMMARY Several file methods were investigated as to their applicability for the primary files; While for particular files the pile file design is adequate; for the more important files the indexed-sequential and direct files are both better and adequate; The indexed-sequential organization is more flexible and generally applicable; Ring-structured files can save storage space but do not have adequate performance; Hiera.rchica1 files; as implemented in MUMPS, have extremely large storage requirements; These could be overcome by restructuring of the hierarchy; This would mean returning to the transformations of the data base model described in Chapter 3 of this thesis and has not been done; The values which were obtained have always assumed average conditions; The margins have been kept sufficiently large so that the combined effect of a very busy day; and an even busier period during such a ... ..."