Version 7.4.4	SEARCH   INDEX   CONTENTS   |   PDF PAGE   PDF BOOKS   |   HOME   UP   PREV   NEXT

This document describes the procedure for defining an Adabas database. It is important for the DBA to understand the information provided for each step before attempting to define a new database.

Defining a new database involves the following steps:

• Step 1 : Estimate the Size of the Database

• Step 2 : Allocate Space

• Step 3 : Format the Space

• Step 4 : Define Database Parameters

After you have completed these steps successfully, you can use the ADACMP and ADALOD utilities to load user files into the database.

Step 1 : Estimate the Size of the Database

Components Required by the Nucleus

The Adabas nucleus requires three database components: Data Storage, an Associator, and a Work area.

Data Storage

The Data Storage component contains the compressed data records of each file in the database.

Associator

The Associator contains elements for each file in the database and for the database as a whole.

For each file in the database, the Associator includes an inverted list, an address converter, and a field definition table (FDT):

• The inverted list, which resolves Adabas search commands and reads records in logical sequence, comprises the normal index (NI) and as many as 14 upper indexes (UI). All of the values for each descriptor in the file are contained in the NI along with a count of the records that contain each value and a list of the ISNs of those records. To increase search efficiency, UI levels are automatically created by Adabas as required, each level to manage the next lower level index. The first level UI, like the NI it manages, contains entries for only one descriptor in each index block. All other UI levels contain entries for all descriptors in each index block. Upper Indexes require a minimal amount of space (two blocks is the minimum).

• The address converter maps the logical identifier of a record (ISN) to the relative Adabas block number (RABN) of the Data Storage block where the record is stored. It comprises a list of RABNs in ISN order; for example, the fifteenth entry in the address converter contains the RABN for ISN 15.

• The field definition table (FDT) defines the logical contents of an Adabas file. It contains the name, level, length, format, and specified options for each field in the file.

For the database as a whole, the Associator includes storage management tables and coupling lists:

• Storage management tables list the Associator and Data Storage blocks that are available for allocation, along with the amount of unused space in each Data Storage block.

• Coupling lists exist for physically coupled files only are used to resolve search commands in which descriptors from more than one file are used.

Work

The Work area stores information in four parts:

• Part 1. Stores data protection information required by the routines for autorestart and autobackout.

• Part 2. Stores intermediate results (ISN lists) of search commands.

• Part 3. Stores final results (ISN lists) of search commands.

• Part 4. Stores data related to distributed transaction processing.

  Note:
  If you have Adabas Transaction Manager Version 7.5 or later installed, Work part 4 of DDWORKR1 is no longer supported. Instead, a second Work dataset, DDWORKR4 is required. DDWORKR4 is a container dataset used for the same purpose as Work part 4 of DDWORKR1, but it can be used in parallel by all members in a cluster. The DDWORKR4 dataset should be allocated and formatted in the normal way, using a block size greater than or equal to DDWORKR1. It should be at least as large as the cluster’s LP parameter of the database or cluster.

Other Components

Sort and Temp Areas

The Adabas utilities ADAINV, ADALOD, and ADAVAL require two additional datasets, SORT and TEMP, for sorting and intermediate storage of data.

The sizes of TEMP and SORT vary according to the utility function to be executed. These datasets can be allocated during the job and then released, or permanent datasets can be allocated and reused.

Logs

Adabas has the following optional logs:

• The command log (CLOG) records information from the control block of each Adabas command that is issued. The CLOG provides an audit trail and can be used for debugging and for monitoring usage of resources. Single, dual, or multiple (2-8) datasets can be used (multiple datasets are recommended ).

• The protection log (PLOG) records before-images and after-images of records and other elements when changes are made to the database. It is used to recover the database (up to the last ET) after restart. Single, dual, or multiple (2-8) datasets can be used (multiple datasets are recommended ).

• The recovery log (RLOG) records additional information that the Adabas Recovery Aid uses to construct a recovery job stream.

Note:
Each CLOG, PLOG, and RLOG dataset is limited to 16,777,215 (x'FFFFFF') blocks/RABNs.

General Space Requirements

The space requirements for the Associator (NI, UI, and AC) and Data Storage are calculated automatically for each file by the ADALOD utility and the ADACMP utility, respectively. If you want to allocate a specific amount of space to a file or estimate the space needed for a file without actually executing these utilities, you can use the formulas provided in this chapter.

If the number and size of the files that will eventually be loaded into the database are not known at the time that the database is established, it is not necessary to allocate a large amount of extra space to the Associator and/or Data Storage, since the space may be increased subsequently by using the ADD or INCREASE function of the ADADBS utility.

The initial allocation for Associator and Data Storage should, however, allow for the loading of all currently planned files in addition to a reasonable amount of database expansion (adding new files or updating existing files).

When estimating the Associator space, the following requirements for the database as a whole must be added to the estimates calculated for each file within the database (normal index, upper indexes, and address converter):

• The first 30 Associator blocks are used by Adabas for storing internal control information. Note that the physical block sizes for Associator, Data Storage, and Work vary from one Adabas component to another and according to the device type on which each component is located.

• Associator blocks equalling five times the value specified by the MAXFILES parameter are reserved by Adabas for file control information. The MAXFILES parameter is set when running the ADADEF utility.

General Procedure for Estimating Space

1. Estimate the following requirements for each file; then add the estimates together for an estimate for the whole database:

   • Associator (normal index, upper indexes, address converter)

   • Data Storage

2. Estimate the following requirements for the database as a whole:

   • Associator (space reserved by Adabas)

     • First 30 blocks for internal control information;

     • (MAXFILES * 5) blocks for file control information (the ADADEF parameter MAXFILES specifies the maximum number of files that can be loaded into the database);

   • Work area; sort area; temp area; logs

Estimation Formulas

The following sections provide formulas for estimating the space that should be allocated to each component.

• Associator, in terms of

  • normal index (NI)

  • upper index (UI)

  • address converter (AC)

• Data Storage

• Work, in terms of

  • part 1 (data protection information)

  • part 2 (intermediate results of search commands)

  • part 3 (ISN lists from search commands)

  • part 4 (data related to two-phase commit processing)

• sort space

Rules of Precedence in the Formulas

The formulas follow the normal rules of precedence; that is, expressions are evaluated in the following order:

1. Elements in parentheses;

2. Multiplication and division operations;

3. Addition and subtraction operations;

4. Left to right (when elements have the same precedence level, the one on the left is evaluated first).

Normal Index (NI)

Use the following formula to estimate the normal index space required for each descriptor in the file:

NIRBYTES = ISNSIZE * AVUQVAL * RECORDS + DESCVALS * (AVLENG + 2)

where

AVUQVAL

AVUQVAL is less than or equal to 1 unless the descriptor is a multiple-value field (MU) or part of a periodic group (PE).

If the descriptor is defined with the NU (null suppression) option, AVUQVAL equals the average number of values per record minus the percentage of records that contain a null value (the field is empty). For example, if each record has one value for the descriptor and 20 per cent of the values are null

AVUQVAL = 1 - 0.2 = 0.8

Similarly, if an MU field has an average of 10 values per record and 20% of the values are null

AVUQVAL = 10 - 2 = 8

AVLENG

If the descriptor field is not defined with the FI (fixed length) option, AVLENG equals the average compressed length of the field, including the length byte. If the descriptor is defined with the FI option, AVLENG equals the standard length of the field.

ISNSIZE * AVUQVAL * RECORDS

ISNSIZE * AVUQVAL * RECORDS represents the space required to store the ISNs. It is the important factor for descriptors that have many duplicate values.

DESCVALS * (AVLENG + 2)

DESCVALS * (AVLENG + 2) represents the space required to store the descriptor values. It is the important factor for descriptors that have a large proportion of unique values.

Convert Bytes to Blocks

Use the following formula to convert bytes to blocks. Round the result up to the next block.

 NIRBLOCKS = NIRBYTES / (ASSOBLKSIZE * (1 - PADFACTOR / 100))

where

Examples

The following examples assume that ASSOR1 is stored on a 3380 device.

Example 1:

Descriptor AA has one value per record and no null values. There are 50 different values for AA in the file. The average compressed length for the values is 3 bytes.

ISNSIZE=3
MAXISN=20000
PADFACTOR=10 (%)

NIRBYTES  = 3 * 1 * 20,000 + 50 * (3 + 2)
                     = 60,000 + 250
                     = 60,250 bytes

NIRBLOCKS  = 60,250 / (2004 * (1 - 0.1))
                        = 33.41
                        = 34 blocks

Example 2:

Descriptor BB has one value per record and no null values. There are 20,000 different values for BB in the file. The average compressed length for the values is 10 bytes.

ISNSIZE=4
MAXISN=20000
PADFACTOR=10 (%)

NIRBYTES  = 4 * 1 * 20,000 + 20,000 * (10 + 2)
                     = 80,000 + 240,000
                     = 320,000 bytes

NIRBLOCKS  = 320,000 / (2004 * (1 - 0.1))
                        = 177.42
                        = 178 blocks

Example 3:

Descriptor CC is a null-suppressed multiple-value (MU) field with an average of 10 occurrences and 3 null values per record. There are approximately 300 different values for CC in the file. The average compressed length for the values is 4 bytes.

ISNSIZE=3
MAXISN=10000
PADFACTOR=5 (%)

NIRBYTES  = 3 * 7 * 10,000 + 300 * (4 + 2)
                     = 210,000 + 1,800
                     = 211,800 bytes

NIRBLOCKS  = 211,800 / (2004 * (1 - 0.05))
                        = 111.25
                        = 112

Example 4:

Descriptor DD is a null-suppressed field contained within a periodic group. DD contains an average of 5 values per record; there is an average of 3 null values per record. There are 10 different values for DD in the file. The average compressed length for the values is 5 bytes.

ISNSIZE=4
MAXISN=10000
PADFACTOR=5 (%)

NIRBYTES  = 4 * 2 * 10,000 + 10 * (5 + 2)
                     = 80,000 + 70
                     = 80,070 bytes

NIRBLOCKS  = 80,070 / (2004 * (1 - 0.05))
                        = 42.06
                        = 43

Upper Index (UI)

Use the following formula to estimate the UI space required for each descriptor in the file:

UIRBYTES = NIRBLOCKS * (AVDESCLEN + ISNSIZE + RABNSIZE + 1)

where

Convert Bytes to Blocks

Use the following formula to convert bytes to blocks. Round the result up to the next block.

UIRBLOCKS = UIRBYTES / (ASSOBLKSIZE * (1 - PADFACTOR / 100))

where

Example

This example assumes that the Associator is stored on a 3380 DASD; therefore, ASSOR1 has 2004 bytes per block.

The NI block requirement for this file is estimated to be 45 blocks. The average compressed length of the values for the descriptor is 3 bytes. The database has 3-byte (24-bit) RABNs; the file has 3-byte ISNs. The ASSOR1 block padding factor is 5 (%).

UIRBYTES  = 45 * (3 + 3 + 3 + 1)
                     = 450

UIRBLOCKS  = 450 / (2004 * (1 - 0.05))
                        = 450 / 1903.8
                        = 0.24
                        = 1 block    ( a minimum of 2 blocks can be allocated for the UI)

Address Converter (AC)

Use the following formula to estimate the address converter space required for the file. Round the result up to the next whole block.

ACBLOCKS = (MAXISN +!) * RABNSIZE / ASSOBLKSIZE

where

Examples

The following examples assume that the Associator is stored on a 3380 DASD; ASSOR1 has 2004 bytes per block.

Example 1:

MAXISN=2000000
RABNSIZE=3

ACBLOCKS  = (2,000,000 * 3) / 2004
                     = 6,000,000 / 2004
                     = 2994.01
                     = 2995 blocks

Example 2:

MAXISN=2000000
RABNSIZE=4

ACBLOCKS  = (2,000,000 * 4) / 2004
                     = 8,000,000 / 2004
                     = 3992.02
                     = 3993 blocks

Data Storage

Use the following formula to estimate the space required for Data Storage. Round the result up to the next whole block.

DATASTORAGE = MAXISN / ((DSBLKSIZE * (1 - (PADFACTOR / 100))) / AVRECLEN)

where

Example

MAXISN = 1000000
Average compressed record length = 50
Model 3380 blocksize for DATA = 4820
Data Storage block padding factor = 5 (%)

DATASTORAGE  = 1,000,000 / ((4820 * (1 - 0.05)) / 50)
                               = 1,000,000 / (4579 / 50)
                               = 1,000,000 / 91
                               = 10,989.01
                               =  10,990 blocks

How Adabas Allocates Work Space

When you allocate the Work datasets, allocate enough space for all parts. The minimum allowable Work space is 300 blocks. Three ADARUN parameters can be used to break up the space into parts 1-4 as follows:

Note:
Work part 4 or DDWORKR1 is no longer supported if you have Adabas Transaction Manager Version 7.5 or later installed. Instead, you should use the DDWORKR4 dataset. DDWORKR4 is a container dataset used for the same purpose as Work part 4, but it can be used in parallel by all members in a cluster. The DDWORKR4 dataset should be allocated and formatted in the normal way, using a block size greater than or equal to DDWORKR1. It should be at least as large as the cluster’s LP parameter of the database or cluster.

• The ADARUN LP parameter specifies the size of Work part 1. The default setting is 1000 blocks; the minimum is 200. A database with little or no updating needs 500-1000 blocks. Work part 1 begins with RABN 1; the last RABN is the value of LP.

• The ADARUN LWKP2 parameter specifies the size of Work part 2. If LWKP2=0 (the default), Adabas calculates the size automatically, using the formula described in Work Part 2: Intermediate Search Results .

Work part 2 begins in the block following the last block of Work part 1; thus, the first RABN of part 2 is given by

1 + LP

• The ADARUN LDTP parameter specifies the size of Work part 4 when ADARUN DTP=RM. If LDTP=0 (the default), the length of Work part 4 is equivalent to the length of Work part 1 (ADARUN LP). If a non-zero value is specified, it must be greater than the value specified for LP. If a smaller value is specified, Adabas changes it to equal the LP value.

  Note:
  Work part 4 of DDWORKR1 is no longer supported if you have Adabas Transaction Manager Version 7.5 or later installed. Instead, you should use the DDWORKR4 container dataset.

Work part 4 begins in the block following the last block of Work part 2; thus, the first RABN of part 4 is given by

1 + LP + LWKP2

• After allocating parts 1, 2, and possibly 4, Adabas allocates the remaining blocks to Work part 3. It is important that you allocate enough space to the DDWORKR1 dataset to leave at least 50 blocks for part 3.

Work Part 1: Data Protection Information

The data protection area for all transactions running in parallel must fit into 1/4 of the Work part 1 (that is, LP) area. Following are general guidelines for determining the proper size for Work part 1:

1. The total Work part 1 size should be four times the estimated size required for a single average transaction in bytes times the maximum number of transactions that run in parallel. This value is then divided by the Work block size (in bytes) minus 200 to convert bytes to blocks.

2. If some transactions are very long, then those transactions alone plus all short transactions executed in parallel should be used to determine the size of a single average transaction.

3. The size of a single average transaction is determined by estimating the average number of updates (modifications, additions, and deletions) per transaction and multiplying that number by the estimated bytes required per update. To this is added space for ET data and for the ET record in bytes.

4. The size required per update is determined by the average compressed record length in bytes times 4 (before image, after image, and DVT space for each) plus 100 bytes for each protection record header (that is, 100 times 4).

A formula that expresses these guidelines is

WK1= ( 4 * TASIZE * TAP ) / ( BLKSIZE - 200 )

where

TASIZE= ( ( ( 4 * AVCRL ) + 400 ) * UPDTA ) + ETDATA + 100

where

Example

If AVCRL = 300 bytes, UPDTA = 4, and ETDATA = 200 bytes, then

TASIZE= ( ( ( 4 * 300 ) + 400 ) * 4 )+ 200 + 100 = 6700 bytes

If TAP = 100 and BLKSIZE = 5492, then

WK1 = ( 4 * 6700 * 100 ) / ( 5492 - 200 ) = 506.46 blocks

Work Part 2: Intermediate Search Results

Use the following formula to estimate the space required for the Work part 2 area. Round the result up to the next whole block.

WORK2 = 22 + 2 * ((4 * RECORDS) / (BLKSIZE - 16))

where

TOPISN - MINISN + 1

Note:
An Adabas internal table requires one byte of storage for each Work part 2 block.

Example

The number of records in the largest file in the database is 500,000. The Work dataset is stored on a 3380 device.

WORK2 = 22 + 2 * ((4 * 500,000) / (5492 - 16))
               = 752.46
               = 753 blocks

Work Part 3: ISN Lists from Search Commands

Adabas allocates to Work part 3 (resultant ISN lists) the Work space remaining after the allocation of the part 1 (data protection information) and part 2 (intermediate results) areas.

The minimum requirement for this area is 50 blocks.

If insufficient space is provided for this area, Adabas may be unable to execute additional search commands until the space currently occupied by ISN lists has been released. Consider the following factors when estimating the space needed for the Work part 3 area:

• The number of concurrent search commands to be processed (each ISN list is stored in a separate block), and the expected size of the resulting ISN lists (each ISN is stored as 4 bytes, regardless of the ISNSIZE specified for the file);

• The number of saved ISN lists resulting from previous search commands with the SAVE ISN LIST option which will be held concurrently;

• The amount of memory which will be required by Adabas as a result (each block allocated to this area requires 4 bytes of memory).

Example

A maximum of 100 search commands with an average of 25 resulting ISNs per command are to be processed concurrently during the session.

Adabas will need 100 blocks in the Work part 3 area.

Work Part 4: Data Related to Distributed Transaction Processing

Note:
Work part 4 of DDWORKR1 is no longer supported if you have Adabas Transaction Manager Version 7.5 or later installed. Instead, you should use the DDWORKR4 dataset. DDWORKR4 is a container dataset used for the same purpose as Work part 4, but it can be used in parallel by all members in a cluster. The DDWORKR4 dataset should be allocated and formatted in the normal way, using a block size greater than or equal to DDWORKR1. It should be at least as large as the cluster’s LP parameter of the database or cluster.

Work part 4 maintains information about some of the global transactions involved in distributed processing. For example, during phase one of the commit process, a global transaction's protection data may be copied from Work part 1 to Work part 4 if it is no longer possible to store the information in Work part 1.

If an overflow of Work part 4 is pending, the nucleus can force a transaction termination. This clears Work part 4 except for transaction IDs (XIDs) and local transaction status information.

The space allocated for Work part 4 must be at least that allocated for Work part 1:

LDTP>=LP

Whether a larger size is needed for Work part 4 depends on the applications running against the database and on the system load.

Because the information maintained in Work part 4 cannot currently be moved to a different area, you can alter the size of Work part 4 between sessions only as follows:

• you can decrease the size of Work part 4 if it was not used at all in the previous session.

• you can increase the size of Work part 4 if it was used in the previous session.

Sort

The following formulas estimate the sort dataset space used for sorting all values of a single descriptor. Multiple descriptors are sorted successively: all values are sorted for the first descriptor, then all values for the second descriptor, and so on. Therefore, estimate the space for the largest possible descriptor sort; that will be enough for all descriptors.

Use the following formula to estimate the space required for the sort area:

DESCSPACE = (AVDESCLEN + (1 + ISNSIZE)) * NUMRECS * AVPEOCCUR * AVMVOCCUR

where

Work Pool Size

Use the following formula to estimate the space required for the work pool:

LWPAVAIL = LWPSIZE - 1216 - (32 * SORTDEVTRKS) - SORTDEVBSIZ

where

Sorted Partial Sequences

To determine the space required for sorted partial sequences, use one of the following calculations. The one to use depends on the AVDESCLEN value (average descriptor length) used to calculate the DESCSPACE value (total descriptor space required).

• If AVDESCLEN is less than 12

LENGSEQ = (LWPAVAIL * (AVDESCLEN + (1 + ISNSIZE))) / 2

where

• If AVDESCLEN is equal to or greater than 12

LENGSEQ = (LWPAVAIL * 2) / 3

where

Device Surfaces

Use the following formula to calculate the number of device surfaces rounded up to the next integer:

SURFACES = (DESCSPACE / LENGSEQ) / SORTDEVTRK

where

Estimated Sort Size

Using the intermediate values calculated for LENGSEQ and SURFACES, compute the estimated sort size as follows:

SORTSIZE = (SURFACES * SORTDEVTRKS * LENGSEQ * 2) / (SORTDEVBSIZ - 4)

where

Number of Descriptors Sorted

Use the following formula to estimate the number of descriptors that can be sorted in the SORTSIZE space calculated in the previous formula (assuming the same descriptor definition that was used when calculating DESCSPACE):

DESCOUNT = SURFACES * SORTDEVTRKS * LENGSEQ / (AVDESCLEN + (1 + ISNSIZE))

where

Step 2 : Allocate Space

1. Use standard operating-system procedures to allocate datasets for the following Adabas components:

   • Required by the Adabas nucleus:

     • Associator (ASSO)

     • Data Storage (DATA)

     • Work area (WORK1)

     • Work area (WORK4), if Adabas Transaction Manager 7.5 or later is installed.

   • Required by some Adabas utilities:

     • sort area (SORT)

     • temp area (TEMP)

   • Optional (but recommended) logs:

     • dual or multiple command log (CLOG)

     • dual or multiple protection log (PLOG)

     • recovery log (RLOG)

     Normally, ASSO, DATA, and WORK are each allocated as a single operating system dataset. However, you can allocate the Associator and Data Storage on up to five separate datasets each; the datasets can be allocated on the same or different device types.

2. To minimize contention and distribute I/O activity more evenly across hardware channels, place the ASSO, DATA, WORK, PLOG, and RLOG datasets on different physical volumes. If only two volumes are available, place ASSO on one volume and DATA and WORK datasets on the second.

   The WORK and PLOG datasets should be on different volumes, since a PLOG I/O operation is always followed by a WORK I/O operation.

   The RLOG dataset should always be placed on a separate device of the same type.

   Disk access time may be considerably reduced by separating TEMP from DATA, and SORT from ASSO. When loading files containing 100,000 records or more, splitting SORT across two volumes reduces disk arm movement.

3. Specify the disk space allocation in the job control (JCL/JCS or VM CONTROL minidisk) of the format utility (ADAFRM). See the Adabas Utilities documentation for specific information and job examples.

Examples

Example 1 : Database Allocation Using Two Volumes

Example 2 : Database Allocation Using Three Volumes

Example 3 : Database Allocation When Loading a Large File

Performance Note

Software AG does not recommend using hardware compression (IDRC) for protection log files. The ADARES utility BACKOUT function will run at least twice as long under OS/390 or z/OS when processing compressed data. Also, the BACKOUT function is not supported for compressed data on VSE/ESA, VM/ESA, or z/VM systems.

Step 3 : Format the Space

Before loading the first file into the database, use the ADAFRM utility to format the ASSO, DATA, and WORK datasets. Refer to the Adabas Utilities documentation for information about the ADAFRM utility.

Format TEMP and SORT before using any Adabas utility that requires them. You can allocate and format temporary datasets and delete them after executing the utility, or allocate and format permanent datasets for repeated use.

Note:
When using the Recovery Aid (including the RLOG), you must catalog all temporary datasets. When running with the Recovery Aid, the general rule is to catalog temporary datasets in jobs that require the Associator datasets.

Format the CLOG, PLOG, and RLOG datasets before starting the first session in which the logging is to be performed.

Step 4 : Define Database Parameters

Once all database components have been physically allocated and formatted, use the ADADEF utility to define database parameters such as database identification, maximum number of files, system file assignment, and so on.

The sizes of the ASSO, DATA, and WORK datasets must be defined with ADADEF DEFINE parameters. Note that defining the sizes to Adabas is different from allocating the space; the datasets must be allocated and formatted before you can define them to Adabas. The sizes of the other datasets are defined to Adabas as follows:

• TEMP and SORT: when you execute the utility that uses them;

• CLOG and PLOG: at the start of a nucleus session, with ADARUN parameters;

• RLOG: when logging begins, using the PREPARE function of the ADARAI utility.

Note:
Each log dataset (CLOG, PLOG, or RLOG) is limited to 16,777,215 (x'FFFFFF') blocks/RABNs.

www.softwareag.com

Copyright © Software AG 1971-2009. All rights reserved.

SEARCH   INDEX   CONTENTS   |   PDF PAGE   PDF BOOKS   |   HOME   UP   PREV   NEXT