This section provides guidelines for defining and tuning Adabas Fastpath parameters to achieve maximum optimization.
Adabas Fastpath optimizes Adabas commands by storing formatted data in a common memory area (Adabas Fastpath buffer area) and returning this data directly to the application without calling the Adabas database. When this is successful, an Adabas command is optimized. The Adabas Fastpath buffer is initialized and managed by the asynchronous buffer manager (AFPABM).
Adabas Fastpath tuning is an ongoing process. Because tuning is partly achieved by changing controls and measuring the results, it is important to decide how you are going to measure the success of Adabas Fastpath in a dynamic production environment. You also need to decide the period over which you will measure the performance based on procedures at your site; for example, many sites have additional weekend and/or month-end processing.
If your environment already measures overall performance, you may decide to use these same measurements for Adabas Fastpath. You may choose to use Adabas Fastpath statistics for overall commands optimized or percentage of commands optimized at the buffer, database, file, or job level. It is important to have a basis for measuring the effect of changes so that you are aware of any decrease in performance.
Adabas Fastpath parameters, which are located in the Adabas Fastpath configuration file, are defined and maintained using Adabas Fastpath Online Services, Parameter Maintenance.
Adabas Fastpath counts the commands optimized and produces other statistics needed to tune its parameters. This information is also recorded on the Adabas Fastpath configuration file. This information can be displayed using the Adabas Online Services functions Buffer Statistics and Buffer History.
The following sections describe the three Adabas Fastpath parameter types: buffer, file, and job. These interact closely in that information from one parameter type may indicate that a parameter of another type may also need to be reviewed.
| Section | Description |
|---|---|
| Buffer Parameters | The Adabas Fastpath buffer is initialized and managed by the asynchronous buffer manager (AFPABM). Each buffer is defined using buffer parameters which can significantly affect the overall performance of Adabas Fastpath. This section tells you how to monitor the buffer statistics and which buffer parameters you may need to modify. |
| File Parameters | File parameters are used to define the databases and files which are to be optimized. This section describes how to select files for optimization, define file optimization parameters, and measure the optimization results. |
| Job Parameters | Each job that uses Adabas issues calls through a link module that includes the appropriate Adabas Fastpath module. This module uses a job parameter located in the Adabas Fastpath configuration file to identify the Adabas Fastpath buffer to be used by this job. This section describes the optimization parameters available at the job level and how to measure their effectiveness. |
The parameters defined for the Adabas Fastpath buffer determine the size and structure of the buffer and have a significant effect on overall performance.
You can use the Adabas Fastpath Online Services to display the main Buffer Performance Measurement screen shown below:
16:32:58 ***** A D A B A S FASTPATH SERVICES 7.4.1 - DEMO ***** 2001-10-05
- General Parameters - F12100MF
Buffer: Global SYSCO33 Node ID: 33
Initial Commands
Main Parameters Value Allocation Peak Usage Ignored
Size(k)..Min: 10240 30720 20480 98%
Maximum Jobs: 64 33 0
RB Length Limit: 2048 0
Set Length Limit: 1024 0
Read-ahead Memory Sizes(k): 4 8 16 32
Other Parameters Value Value
Direct Set Expansiom: 8 Set Concurrency: 2
Freespace Index Entries: 1024 Average Item Size: 256
Log every n minutes: 60 Keep for n days: 30
Autorestart: Y Restart every n Hrs: 0
Direct Access: ON Read-ahead Processing: ON
FIND SX/L1: ON READ PHYSICAL L2: ON
READ LOGICAL L3: ON HISTOGRAM L9: ON
Command ==>
Enter-PF1---PF2---PF3---PF4---PF5---PF6---PF7---PF8---PF9---PF10--PF11--PF12
Help Exit Serv Excl Gen Node Menu |
Each of the main buffer parameters shown on this screen is described in the table below.
The following information is provided:
the parameter value currently defined;
the initial memory allocation;
the peak usage (which includes extra areas allocated); and
the commands ignored because of an area shortage or limit having been exceeded.
| Parameter | Description |
|---|---|
| Size |
The buffer size (minimum, defined, initial allocation, peak usage). In this example, the minimum is defined as 10 megabytes and the defined size is 30 megabytes. The asynchronous buffer manager will request 30 megabytes of contiguous common memory, and if this is not available, a minimum of 10 megabytes will be requested. If the minimum is not available, the buffer cannot be initialized. In the example, an initial allocation of 20 megabytes has been made and the peak usage is currently 98%, that is the buffer is in effect fully used. Any value above 97% is regarded as 100% usage (there is usually a small amount of unused residue because of fragmentation). A fully used buffer does not necessarily mean that the size needs to be increased. Other indicators must also be considered; for instance, low direct access optimization results. It is not possible to define a general formula for increasing this parameter. The best approach is to redefine the size and then measure the overall effect on performance. If performance is improved, you can repeat the process until there is no further performance gain. |
| Maximum Jobs |
The maximum number of jobs permitted for the buffer. This value is used to preallocate the number of areas needed for all Adabas Fastpath jobs. Each Adabas Fastpath job is allocated a job area when it is started. When a job ends, the area is freed and can be reused. Peak usage is the highest number of areas which have been allocated. This parameter value should be increased if
|
| RB Length Limit |
The maximum length of a record buffer for direct access optimization processing. Commands ignored because of the limit are also shown for this parameter. In the example, 5,476 commands were ignored. This indicates that this parameter value could be increased to achieve better optimization. |
| Set Length Limit |
The maximum length of a set used for direct access optimization processing. A set is made up of the Adabas format buffer (FB) and search buffer (SB) plus an extra 12 bytes. Commands ignored because of the limit are also shown for this parameter. If a large number of commands are being ignored, this parameter value could be increased to achieve better optimization. |
| Read-Ahead Memory Sizes |
The limits for the amount of memory that can be used for read-ahead optimization. |
| Other Parameters | For information on all other parameters, see the section Adabas Fastpath Parameters. |
The section describes how to select Adabas files for optimization, and how to define individual file parameters for those files which have been selected for optimization.
Adabas Fastpath provides the AFPLOOK tool for sampling an Adabas database and reporting on the suitability of individual files for Adabas Fastpath optimization. This section provides general guidelines regarding the contents and usage of the information provided by AFPLOOK.
Note:
For additional documentation on AFPLOOK, see
AFPLOOK and
Using AFPLOOK from Adabas Online
Services.
The AFPLOOK samples provided in this section were created using Adabas Online Services and are based on the following conditions:
the database is sampled in its entirety with no focus on particular files or jobs,
the sampling session extends over a period long enough to be representative of the normal workload,
any other databases in use are sampled and analyzed in the same way so that the whole system is optimized, not just a part of it.
The sample presents a realistic use of AFPLOOK with the goal of understanding the general workload presented to the database. The screens included in this section are taken from an in progress sampling session. Therefore, the numbers contained in the detailed screens do not correlate with the numbers in the summary screens. If necessary, you can "freeze" a sample so that detail and summary numbers do correlate by requesting the sample to stop. Sampling stops, but the numbers are retained for analysis purposes. You can then release the sample and continue the session.
AFPLOOK sampling results provide the information needed to initially define optimization parameters. You may need to adjust these parameter settings after observing the results of subsequent processing.
AFPLOOK reports on the commands processed by the database. If you are already optimizing with Adabas Fastpath, AFPLOOK does not account for the commands which are already optimized, since they are not sent to the database.
For files where optimization has been defined, AFPLOOK only counts commands that either have not been qualified or have failed optimization. Optimized direct access commands do not appear on the report at all, even in file totals. Optimized read-ahead commands do not appear either, but the command used to prime read-ahead is counted in the prefetch counter, indicating that prefetch/multifetch has already been set for these commands.
Files where no optimization has been defined show accurate figures except where a default file 0 has been defined for general read-ahead for the database. In this case, any file can be affected and will ignore optimized commands and show counts in the prefetch counter.
When interpreting a file sample, it is important to remember that the AFPLOOK view can be seriously affected when Adabas Fastpath is active. Because optimized commands are not seen by the sample, even the file totals and relative percentages are affected. Therefore, do not be misled into thinking that the file parameters defined are not needed when in fact, they are performing well.
The Potential Optimization Summary screen provides potential optimization information for an Adabas database.
12:33:10 ***** A D A B A S FASTPATH SERVICES 7.4.1 ***** 2001-09-05
- Afplook - Summary for DB 50 - F16300MA
Afplook Started Sep 5,2001 at 12:30 by UKDEV1
----------------------- Potential Optimization Summary -----------------------
Sampled Commands Potential Optimization
Sampled File Commands 57960 (75%) <=== Sequential: 2797 ( 5%)
Direct Access: 30597 (53%)
RCs: 22194 (38%)
Excluded Commands 19115 (25%)
Totals 77075 (100%) 55588 (72%)
--------------------------------------------------------------------------------
Command ==>
Enter-PF1---PF2---PF3---PF4---PF5---PF6---PF7---PF8---PF9---PF10--PF11--PF12
Help Exit Refr File More Menu |
In the sample screen, the field Sampled File Commands shows a total of 77,075 commands processed by Adabas during the sample period of which 75% are considered to have optimization potential. The field Excluded Commands shows that 25% are not considered to have optimization potential.
The full optimization potential is generally not fully achievable. Experience shows that approximately 50% of the potential is most likely to be achieved. In this example, 30-40% optimization can be expected for this database which would result in an approximate Adabas CPU usage reduction of approximately 20%.
You can use PF11 to display more information on Excluded Commands.
12:33:41 ***** A D A B A S FASTPATH SERVICES 7.4.1 ***** 2001-09-05
- Afplook - Summary for DB 50 - F16300MB
Afplook Started Sep 5,2001 at 12:30 by UKDEV1
----------------------------- Command Analysis -------------------------------
Rejected Commands:
Max.Users Exceeded:
Max.CIDs Exceeded:
Max.Files Exceeded:
Excluded Commands:
Bad Commands:
Non-File Commands: 431
Non-File RCs: 1718
Excluded File Commands: 6015
Update Commands: 10951
Already Prefetched: 19115 (25%)
Sampled File Commands: 57960 (75%)
All Commands Seen: 77075 (100%)
------- Parameters and High Water Marks Reused User Areas:
Max.Files: 64 Files Needed: 28 Max. DE: 32 Overflows:
Max.Users: 100 Peak Users: 1 Max.CID: 10 Peak CIDs: 5
Command ==>
Enter-PF1---PF2---PF3---PF4---PF5---PF6---PF7---PF8---PF9---PF10--PF11--PF12
Help Exit Refr File Prev Menu |
The section Rejected Commands shows the number of commands which were rejected as a result of the maximum number of users, maximun number of Command IDs, or maximum number of files being exceeded.
The section Excluded Commands shows the number of commands excluded because of the following reasons:
Bad Commands: commands not recognized or invalid.
Non-File Commands: commands which are not related to file processing.
For example, ET,BT,OP,CL.
Non-File RC Commands: Release
Command ID commands which are not related to file processing.
Excluded File Commands: access commands which cannot be optimized.
For example, L4,L5,L6,S4.
Updated Commands: update commands which cannot be optimized. For
example, A1, E1, N1.
Already Prefetched: prefetching may exclude a significant number of commands that otherwise could be optimized. If so, it may be worth running the sample without prefetching to achieve a more realistic result.
The parameters used to control the AFPLOOK sample are displayed at the bottom of the screen. The limits set for these parameters may result in the rejection of commands during sampling. Rejections, if any, are also displayed on this screen. If the number of commands rejected is significant, you should increase the value for the relevant parameter and then execute the sample again.
You can use PF10 on the Database Summary screen to invoke the File Summary screen to determine the most appropriate files for optimization.
Note:
The list of files usually extends to more than one screen. Because
of space restrictions, some files have been removed from this example.
12:31:26 ***** A D A B A S FASTPATH SERVICES 7.4.1 ***** 2001-09-05
Afplook - Files Summary for DB 50 - F16200MA
Started: Sep 5,2001 at: 12:30 by: UKDEV1
C Fnr Total Direct Access RCs Sequential
_ 50 3282 ( 3%) 1641 (50%) 1641 (50%)
_ 52 2428 ( 2%) 1214 (50%) 1214 (50%)
_ 53 3144 ( 3%) 1564 (50%) 1564 (50%)
_ 55 1644 ( 1%) 822 (50%) 822 (50%)
_ 64 821 ( 1%) 821
_ 87 1640 ( 1%) 820 (50%) 820 (50%)
_ 135 12432 (10%) 7102 (57%) 5330 (43%)
X 137 51540 (43%) 28455 (55%) 16876 (33%) 3547 ( 7%)
_ 139 5324 ( 4%) 2662 (50%) 2662 (50%)
_ 143 19514 (16%) 8870 (45%) 7983 (41%) 1774 ( 9%)
_ 144 3551 ( 3%) 1776 (50%) 1775 (50%)
_ 191 3226 ( 3%) 1378 (43%) 908 (28%) 470 (15%)
_ 250 9432 ( 8%) 5623 (60%) 3809 (40%)
Total 122202 (100%)
Mark to Display File Details
Command ==>
Enter-PF1---PF2---PF3---PF4---PF5---PF6---PF7---PF8---PF9---PF10--PF11--PF12
Help Exit Refr Sumry |
The following information is provided for each file:
| Field | Description |
|---|---|
| Total | The total number of commands processed for the file. The accompanying percentage can be used to identify quickly the busiest files, especially when the file summary is spread across several screens. |
| Direct Access | The number of direct access commands processed for the file. |
| RCs | The number of RC
commands processed for the file. The Adabas
RCcommand releases one or more Command IDs currently
assigned to a user. See the Adabas Command Reference
documentation for more information. When a direct access command is optimized,
it is often possible to optimize its matching RC
command.
|
| Sequential | The number of sequential access commands processed for the file. |
In the example, file 137 is by far the busiest file (43% of database commands are for this file), and 95% of the sample for file 137 has potential optimization (55%+33%+7%).
By marking File 137, further details can be displayed for the purpose of selecting the appropriate optimization parameters for this file.
12:32:21 ***** A D A B A S FASTPATH SERVICES 7.4.1 ***** 2001-09-05
- Details for DB 50 File 137 - F16210MA
Started: Sep 5,2001 at: 12:30 by: UKDEV1
CC DE Direct Acc RCs Sequential Sequences Non-Qualifying
L3 S5 23902 20508 6752 394 Updates 32
S1 S5 54584 27093 Inserts 0
L3 S1 23670 16889 6781 6781 Deletes 0
L1 -- 6748
L9 TA 13032 13032 Prefetch 0
L9 S5 9326 16478 7896 Excluded 32
Spare 0
No CID 0
Max.CIDs 0
Bad Rsp. 0
Max.Rec.Buff.Len
Total 131262 77522 30011 Direct Acc 69
Total Commands for File 253970 Sequential 63
Command ==>
Enter-PF1---PF2---PF3---PF4---PF5---PF6---PF7---PF8---PF9---PF10--PF11--PF12
Help Exit Refr Parms File Menu |
This screen shows the commands that can be optimized and summarizes those that cannot.
The following information is provided:
| Field | Description |
|---|---|
| CC DE | The Adabas command and descriptor pair for which information is being provided. |
| Direct Acc | The number of direct access commands which have optimization potential. |
| RCs | When a direct access command is optimized, it is
often possible to optimize its matching RC command.
This column indicates the potential number of RC
commands which can be optimized.
|
| Sequential Sequences |
As a general rule, read-ahead optimization is only appropriate for descriptors with at least a 2 to 1 ratio of sequential commands to sequences. If the ratio is smaller, over-reads could occur if the descriptor is optimized. In the example, for the command/descriptor pair
The example also shows, for the command/descriptor pair
|
| Non-Qualifying | Commands that cannot be optimized. Also shown are the record buffer lengths used for direct access and sequential views. |
Before defining the optimization parameters, it may be useful to have a copy of the AFPLOOK file detail statistics for the selected files to refer to when you are monitoring the results.
PF5 can now be used to define the optimization parameters as described in the next section.
This section describes the definition of file optimization parameters using various file examples:
As shown on the example screen below, for file number 135, all
L3 and S1 commands
potentially qualify for direct access optimization.
12:32:21 ***** A D A B A S FASTPATH SERVICES 7.4.1 ***** 2001-09-05
- Details for DB 50 File 135 - F16210MA
Started: Sep 5,2001 at: 12:30 by: UKDEV1
CC DE Direct Acc RCs Sequential Sequences Non-Qualifying
S1 AA 10188 10188 Updates 6749
L3 AA 3396 3396 Inserts 0
S1 A1 6748 6748 Deletes 0
S1 BS 6717
Prefetch: 0
Excluded: 0
Spare : 0
No.CID : 0
Max.CIDs: 0
Bad Rsp.: 0
Max.Rec.Buff.Len
Total 27049 20332 Direct Acc 648
Total Commands for File 47381 Sequential
Command ==>
Enter-PF1---PF2---PF3---PF4---PF5---PF6---PF7---PF8---PF9---PF10--PF11--PF12
Help Exit Refr Parms File Menu |
Direct access optimization is therefore defined as ON for
L3 and S1 commands
without any field level (descriptor) overrides:
On/Off Except for: Field level overrides
L1 Get ISN OFF
L3 Read Logical ON __ __ __ __ __ __ __
L9 Histogram OFF __ __ __ __ __ __ __
S1 Find Records ON __ __ __ __ __ __ __
S2 Find Sorted OFF __ __ __ __ __ __ __ |
Defining commands for optimization without descriptor overrides allows the automatic inclusion of other descriptors that may not be visible in this example or may be used to access the file in the future. This method is a good choice for
a reference type file where you anticipate direct access optimization; or
a file with more than seven descriptors, all with good optimization potential, for a command.
In the next example, direct access optimization is to be performed
only for L3 commands which use the field AA as
primary descriptor and for S1 commands which use the
field AA, A1 or BS as primary descriptor. This method is a good choice for a
file with many descriptors and for which most of the database access is
performed using specific descriptors.
Note:
Monitoring the performance of file optimization at descriptor level
is the same whether generic command definitions are used or not. However, the
result of monitoring may require that you change from generic to explicit
descriptor definitions, or that you explicitly exclude certain
descriptors.
On/Off Except for: Field level overrides
L1 Get ISN OFF
L3 Read Logical OFF AA __ __ __ __ __ __
L9 Histogram OFF __ __ __ __ __ __ __
S1 Find Records OFF AA A1 BS __ __ __ __
S2 Find Sorted OFF __ __ __ __ __ __ __ |
In the example below, for file 143, there is read-ahead optimization
for the command/descriptor pairs L3/S2, L3/S3 and
L3/S4, as shown by the values in the Sequential and
Sequences fields.
12:32:21 ***** A D A B A S FASTPATH SERVICES 7.4.1 ***** 2001-09-05
- Details for DB 50 File 143 - F16210MA
Started: Sep 5,2001 at: 12:30 by: UKDEV1
CC DE Direct Acc RCs Sequential Sequences Non-Qualifying
S1 S3 10142 10142 Updates 16921
L3 S2 23701 20311 6780 3390 Inserts 0
L3 S3 500 306971 654 Deletes 0
L3 S4 12403 12401 12401
L3 S7 134 Prefetch 0
L3 S9 99 Excluded 16921
Spare : 0
No.CID : 0
Max.CIDs: 0
Bad Rsp.: 0
Max.Rec.Buff.Len
Total 46979 30453 326152 Direct Acc 195
Total Commands for File 406974 Sequential 195
Command ==>
Enter-PF1---PF2---PF3---PF4---PF5---PF6---PF7---PF8---PF9---PF10--PF11--PF12
Help Exit Refr Parms File Menu |
In that the L3/S3 pair has a relatively low
volume of commands, only the pairs L3/S2 and
L3/S4 are defined for optimization.
On/Off Except for: Field level overrides
L1 Get ISN OFF
L3 Read Logical OFF S2 S4 __ __ __ __
L9 Histogram OFF __ __ __ __ __ __ __
S1 Find Records OFF S3 __ __ __ __ __ __
S2 Find Sorted OFF __ __ __ __ __ __ __ |
However, in the case of read-ahead parameters, descriptor
S4 should not be defined because the number of
sequential commands is equal to the number of sequences which means that no
real read-ahead capability exists. Therefore, it is necessary to explicitly
define descriptor overrides for L3 commands:
Note:
If descriptor S4 were defined for
optimization (or implicitly allowed), a large number of over-reads would
probably result.
On/Off Except for: Field level overrides
L1 Read by ISN OFF
L2 Read Physical OFF
L3 Read Logical OFF S2 S3 __ __ __ __ __
L9 Histogram OFF __ __ __ __ __ __ __
S1 Find Records OFF __ __ __ __ __ __ __
S2 Find Sorted OFF __ __ __ __ __ __ __
S8 Process ISN List OFF
S9 Sort ISN List OFF |
Read-ahead parameters can also be defined by setting
L3 command optimization ON and by specifying the
descriptors which are not to be optimized.
On/Off Except for: Field level overrides
L1 Read by ISN OFF
L2 Read Physical OFF
L3 Read Logical ON S4 __ __ __ __ __ __
L9 Histogram OFF __ __ __ __ __ __ __
S1 Find Records OFF __ __ __ __ __ __ __
S2 Find Sorted OFF __ __ __ __ __ __ __
S8 Process ISN List OFF
S9 Sort ISN List OFF |
A file parameter defined with a file number value of 0 indicates a default definition for all files in the database that do not have an explicit file parameter definition. This can be useful for:
defining read-ahead optimization, but no descriptor overrides, for all affected files.
a database with many files for which only read-ahead optimization is appropriate.
Note:
Direct access optimization is not possible by default.
All statistics for the affected files are shown under the default. This makes it difficult to monitor individual files. For example, if over-reads are high, you may not be able to identify the file that is causing the problem.
If you do identify the file causing over-reads, you can switch off read-ahead optimization for the file by defining a file parameter for that file.
The section describes how to monitor the performance of the currently defined optimization parameters.
You can use Adabas Fastpath Online Services to display a summary of the file optimization performance for a given file.
The following example shows the summary for file 137 in database 50.
10:33:26 ******* A D A B A S FASTPATH SERVICES 7.4.1 ******* 2001-09-05
- File Details from Global SYSCO33 Node Id 33 - F12210MB
Database: 50 SVC 240 File: 137
---------------------------- Optimization Summary ----------------------------
Access Type Attempts (87%) <----------Successes (65%)
Read-ahead: 2,918 (32%) 1,765 (60%)
Direct: 3,914 (43%) 2,313 (59%)
Rcs: 1,170 (12%) 1,170 (29%)
Excluded: 1,062 (11%)
Total: 9,064 (100%) 5,248 (65%)
------------------------------------------------------------------------------
Updates: Over-reads: 668 (37%)
Set Limit(k): 1024 Set Expansion: 8 Update Sensitivity: R
Set Id Length Limit: 1024 RB Length Limit: 2048 Set Concurrency : 2
Status: 00 D4 Last RC: Next Event: 0:00 Addr: 099F29C0
Command ==>
Enter-PF1---PF2---PF3---PF4---PF5---PF6---PF7---PF8---PF9---PF10--PF11--PF12
Help Exit Refr Excl More Zero Node Menu |
With an overall optimization success rate of 65%, both direct access and read-ahead optimization can be considered to be successful.
Nonetheless, it may be possible to obtain a higher optimization percentage by reviewing more detailed information which is provided using PF6.
10:31:15 ***** A D A B A S FASTPATH SERVICES 7.4.1 ***** 2001-09-05
- File Details from Global SYSCO33 Node Id 33 - F12210MF
Database: 50 SVC 240 File: 137
+---------------------------------------+
Excluded:- Reasons Number | Direct not Optimized Number |
Misc/temp off switch | Repo. size overflow |
User area shortage | Concurrency clash 3 |
Non-AFP CB settings | Startup/set off/rej 442 |
CID area shortage | DA set not found 731 |
M/fetch already used | DA data not in set 425 |
SetId length exceed | RBL exceeds maximum |
Desc. not qualified 1023 | |
FB/SB not recognized +---------------------------------------+
Unknown status 1 | Read-ahead Not Optimized |
Sequence detection 38 | Memory shortage 75 |
| Read-ahead primer 1078 |
| Sx command ignored |
| |
+---------------------------------------+
Command ==>
Enter-PF1---PF2---PF3---PF4---PF5---PF6---PF7---PF8---PF9---PF10--PF11--PF12
Help Exit Refr Gen Zero Node Menu |
The information provided on this screen is described in the following sections:
The first File Detail screen for file 137 shows that 11% of the commands are excluded from optimization.
The second File Detail screen provides the specific reasons for
exclusion. It shows that 1023 commands were excluded because the descriptor was
not qualified. By referring back to the saved AFPLOOK output for file 137, on
which the parameter selection was based, it can be seen that
L3 commands with descriptor S1 were not optimized,
which explains the large number of excluded commands.
The other numbers in the Excluded Reasons column are small and are consistent with expectations for a normal environment.
Note:
The information on the right hand side of the screen is discussed in
subsequent sections. In addition, a description of each exclusion category is
provided in section Excluded Command
Summary, together with information about the parameters that affect
each reason.
The first File Details screen for file 137 shows that direct access
optimization is 59% successful and an additional 29% of RC
commands are optimized.
The second File Detail screen provides the specific reasons why some direct access commands are not optimized.
10:31:15 ***** A D A B A S FASTPATH SERVICES 7.4.1 ***** 2001-09-05
- File Details from Global SYSCO33 Node Id 33 - F12210MF
Database: 50 SVC 240 File: 137
+---------------------------------------+
Excluded:- Reasons Number | Direct not Optimized Number |
Misc/temp off switch | Repo. size overflow |
User area shortage | Concurrency clash 3 |
Non-AFP CB settings | Startup/set off/rej 442 |
CID area shortage | DA set not found 731 |
M/fetch already used | DA data not in set 425 |
SetId length exceed | RBL exceeds maximum |
Desc. not qualified 1023 | |
FB/SB not recognized +---------------------------------------+
Unknown status 1 | Read-ahead Not Optimized |
Sequence detection 38 | Memory shortage 75 |
| Read-ahead primer 1078 |
| Sx command ignored |
| |
+---------------------------------------+
Command ==>
Enter-PF1---PF2---PF3---PF4---PF5---PF6---PF7---PF8---PF9---PF10--PF11--PF12
Help Exit Refr Gen Zero Node Menu
|
The following fields are provided:
| Field | Description |
|---|---|
| Repo. size overflow |
The number of commands that could not be optimized because the reposition area in the Adabas Fastpath user area was not large enough. |
| Concurrency clash |
The number of commands that could not be optimized because
|
| Startup/set off/rej | This field normally shows a count of the commands not optimized during the Adabas Fastpath knowledge base building phase. Once the initial knowledge base has been built, this number should remain relatively static. |
| DA set not found | The number of times a required set built and used by Adabas Fastpath was not found. This number is high until the set areas are established. If the number continues to increase, there is probably a general shortage of memory in the Adabas Fastpath buffer. See also Direct Access Sets. |
| Da data not in set |
The number of sets built for which data has not been captured. This number will be high during the knowledge base building phase. This count continues to increase; however, if the rate of increase continues to be high, either
The section Direct Access Sets describes additional checks that can be made if this field value appears to be too high. |
| RBL exceeds maximum | The number of times that the maximum value set
for the record buffer length has been exceeded. This value is set by the
parameter RB Length Limit and restricts the size of
buffer used for evaluating data for direct access optimization.
|
Adabas Fastpath maintains direct access data by database and file in a set. Each set has a unique data profile. For any given Adabas command/descriptor pair defined for optimization, multiple sets can result each with a different profile.
The screen example below shows set information for each command/descriptor pair for file 137.
This screen provides the information necessary to decide whether or not the currently defined file optimization parameters defined have been successful. A line is shown for each command/descriptor pair for which direct access optimization sets have been built. Each line shows the Adabas command, the primary descriptor and the number of direct access optimization sets that have been built for this combination.
10:38:27 ***** A D A B A S FASTPATH SERVICES 7.4.1 ***** 2001-09-05
- Set Summary from Global SYSCO33 Node Id 33 - F12220MA
Database: 50 SVC 240 File: 137
Number of Items.... Memory (k).... Efficiency Ratings
C CC De Sets In Use Peak Optimized In Use Peak Set Memory Growth
_ L1 1 49 56 11 44 44 0 1 0.01
_ L3 S1 1 51 57 1557 5 6 9 311 2.99
_ L3 S5 2 220 234 1887 30 32 1 62 0.59
_ L9 S5 1 20 24 404 1 2 9 404 3.88
_ L9 TA 1 120 124 1427 3 3 9 427 4.10
_ S1 S5 1 2 8 758 7 7 9 108 0.90
Mark with D(etail),F(ull Set List)
Command ==>
Enter-PF1---PF2---PF3---PF4---PF5---PF6---PF7---PF8---PF9---PF10--PF11--PF12
Help Exit Refr Parms Look Cntl Node Menu |
In this example, the number of sets is low. In a heavy production environment, it is possible for some command/descriptors pairs to have a large number of sets. For each set, the information described in the table below is provided. The values provided for each command/descriptor pair are an accumulation of the individual sets, except for those values provided in the column Efficiency Ratings, which are recalculated from various totals of the individual sets.
| Field | Description |
|---|---|
| Number of Items |
In Use: the number of items (data records) currently in the
sets. |
| Memory |
In Use: the amount of memory allocated from the Adabas
Fastpath buffer currently in use for the sets. |
| Efficiency Ratings |
The efficiency ratings for set, memory and growth. These ratings are calculated from the above values to obtain measurements that can be used to judge the performance of the command/descriptor combinations. |
| Efficency Rating (Set) |
The effort needed to maintain the sets (adding, deleting, overwriting) against the (nominal) effort saved by optimization.
|
| Efficiency Rating (Memory) |
The number of optimizations against the memory used. The higher the rating, the better the value of the sets in terms of memory usage. |
| Efficiency Rating (Growth) |
Growth efficiency puts the memory value into perspective by relating it to the average for the whole buffer. A rating of 1.00 is average. |
In the example screen, the L1 command
(descriptors are not applicable for L1s) has a set
efficiency of 0. The memory value is also low. Therefore, optimization for this
command should be removed unless the performance is likely to change.
In the File Set Summary screen, the L3
command with descriptor S5 shows a set efficiency of
1 and a memory value that is approximately half of the average for the buffer.
By using the D option next to the L3/S5 pair, more
detailed information for the sets used by this pair can be displayed:
10:38:27 ***** A D A B A S FASTPATH SERVICES 7.4.1 ***** 2001-09-05
- Set Summary from Global SYSCO33 Node Id 33 - F12220MB
Database: 50 SVC 240 File: 137
Command: L3 Primary Descriptor: S5 Number of Sets: 2
Efficiency Ratings: Set: 1 Memory: 62 Growth: 0.59
Items - Optimized: 1887 Added: 220 Collisions: 6
Current: 220 Peak: 234 Free: 14
Removed: RLU: Overwrites: 1433
Set Resets:
Memory- Used(k): 30 Peak(k): 32 Free(k): 1
Command ==>
Enter-PF1---PF2---PF3---PF4---PF5---PF6---PF7---PF8---PF9---PF10--PF11--PF12
Help Exit Refr Parms Look Node Menu |
This screen shows the totals for the two sets that have been created
for the command L3 descriptor
S5 combination.
In addition to the fields previously described on the preceding screen, this detail screen shows the following fields:
| Field | Description |
|---|---|
| Collisions | The number of times a set was inaccessible because concurrency for the set was exhausted. |
| Overwrites |
The number of overwrites which have occurred for the sets. On overwrite occurs when Adabas Fastpath tries to insert a new data record but finds that there are no free item areas available for the set. When this happens, the set is marked for expansion and an aging algorithm is used to determine if a record is suitable for deletion. If so, this record is overwritten by the new record. In this example, there is an unusually high number of overwrites relative to the number of items added. It is common for sets to have overwrites, especially when first being built, but these usually decrease as a set matures. This summary in the example shows a typical profile of sets with a low data reusage ratio. The number of optimizations only minimally exceeds the total of Added and Overwrites, which often indicates that a data record is added to the set but then only optimized once or twice. In this type of set, data records can continue to be inserted and memory can continue to be requested until all the buffer memory is used. Optimization continues to be based on the latest record inserted, so the oldest data is no longer referenced and is therefore wasted. Adabas Fastpath recognizes the pattern of behavior and, because the growth rate is below average, restricts the set expansion. This properly causes the sets to overwrite old data rather than continually insert additional data. In some cases, it may be helpful to look at the various individual direct access sets that compose the set summary for a command/descriptor combination. |
| Set Resets | The number of times a complete set was emptied. |
In the File Set Summary screen, entering the F option next to the
L3/S5 pair invokes the Command Descriptor Set List
screen:
10:39:56 ***** A D A B A S FASTPATH SERVICES 7.4.1 ***** 2001-09-05
- Set Summary from Global SYSCO33 Node Id 33 - F12230MA
Database: 50 SVC 240 File: 137 CC: L3 De: S5
Number of Items...... Memory (k)..... Efficiency Ratings
C SSID In Use Peak Optimized In Use Peak Set Memory Growth
_ 8 218 226 1489 30 31 1 49 0.47
_ 6 2 8 398 9 398 3.82
Mark with D(etail),S(start/stop)
Command ==>
Enter-PF1---PF2---PF3---PF4---PF5---PF6---PF7---PF8---PF9---PF10--PF11--PF12
Help Exit Refr Cntl Node Menu |
This screen shows different behavior profiles for the two sets:
The first set has a higher volume but low efficiency
The second set has low volume but high efficiency.
Although together the sets show a positive contribution to the overall Adabas Fastpath efficiency, the sets do use 30k of memory from the Adabas Fastpath buffer. In other examples, the memory involved could be several megabytes. If memory shortage is causing lost optimization for other databases/files/sets or read-ahead, this command/descriptor pair could be removed from optimization as part of an overall strategy to recover allocated memory.
The file parameters defining optimization can only be defined down to the command/descriptor level. While it is possible to stop and start individual sets, this is only useful on a short-term basis because the set is restarted every time the buffer, database, or file is restarted.
The Set Details screen can be displayed for reference while considering additional options for controlling individual sets.
10:40:31 ***** A D A B A S FASTPATH SERVICES 7.4.1 ***** 2001-09-05
- Set Detail from Global SYSCO33 Node Id 33 - F12230MB
Database: 50 SVC 240 File: 137
Command: L3 Primary Descriptor: S5 SSID: 8
Efficiency Ratings: Set: 9 Memory: 398 Growth: 3.82
Items - Optimized: 1489 Added: 218 Collisions: 6
Current: 218 Peak: 226 Free: 8
Removed: RLU: Overwrites: 1433
Set Resets:
Memory - Used(k): 30 Peak(k): 31 Free(k): 1
Status: 00 C8 0E 08 Expansion: 8 Acc.Mask Entries 2
Used FB: 54 RB: 70 Indexes: 3 Address: 09A00234
Lengths: SB: 9 VB: 20 Code: Lock: 00000000
SetID Length: 75
Reject:
SB AE,020,A.
FB AA,008,A,AC,020,A,AE,020,A,AD,020,A,AF,001,A,AG,001,A.
Command ==>
Enter-PF1---PF2---PF3---PF4---PF5---PF6---PF7---PF8---PF9---PF10--PF11--PF12
Help Exit Refr Conc Menu
|
For some areas, you can set general buffer or file parameters that have an effect on individual sets. The parameters apply to all the sets within a file:
| Buffer/File Parameter | Description |
|---|---|
| Set Limit (k) | Sets the maximum memory allocation for each set within the buffer or file. If a set uses a lot of memory but you want to continue optimizing the command/descriptor pair, you can restrict the file parameter to limit the maximum memory allocated to a set within the file. |
| Set ID Length Limit | Defines the maximum length of the unique set identifier. A set identifier comprises 12 bytes control informatin plus the Adabas search and format buffers. By reducing the value of this parameter at the file level, you can prevent optimization on certain sets within a file. |
| RB Length Limit | Sets the maximum record buffer length for which sets will be built. The Set Details screen shows the actual record buffer length used (in the case of variable length sets, it is the maximum RB length used so far). By restricting the value of this parameter at the file level, you can prevent optimization on certain sets within a file. |
The first File Details screen for file 137 shows a successful read-ahead optimization percentage of 60%.
The section Read-Ahead Not Optimization on the second File Details screen provides further information as to why some commands could not be optimized.
10:31:15 ***** A D A B A S FASTPATH SERVICES 7.4.1 ***** 2001-09-05
- File Details from Global SYSCO33 Node Id 33 - F12210MF
Database: 50 SVC 240 File: 137
+---------------------------------------+
Excluded:- Reasons Number | Direct not Optimized Number |
Misc/temp off switch | Repo. size overflow |
User area shortage | Concurrency clash 3 |
Non-AFP CB settings | Startup/set off/rej 442 |
CID area shortage | DA set not found 731 |
M/fetch already used | DA data not in set 425 |
SetId length exceed | RBL exceeds maximum |
Desc. not qualified 1023 | |
FB/SB not recognized +---------------------------------------+
Unknown status 1 | Read-ahead Not Optimized |
Sequence detection 38 | Memory shortage 75 |
| Read-ahead primer 1078 |
| Sx command ignored |
| |
+---------------------------------------+
Command ==>
Enter-PF1---PF2---PF3---PF4---PF5---PF6---PF7---PF8---PF9---PF10--PF11--PF12
Help Exit Refr Gen Zero Node Menu
|
| Field | Description |
|---|---|
| Memory shortage |
The number of commands that were not optimized because of a memory shortage. This occurs when the job does not have enough local storage available to support all the read-ahead requirements. |
| Read-ahead primer |
The number of commands used to perform the multifetch function required for read-ahead processing. Such commands are sent to Adabas and are not optimized. Adabas Fastpath read-ahead uses the Adabas multifetch function to pass back one or more records in addition to the record requested by the user application. The additional records are stored until the user application requests them one-by-one. Read-ahead memory areas have a fixed size with up to four different size categories. Adabas Fastpath uses the nearest larger area size if possible. The parameter Read-Ahead Memory can be used to define the sizes of read-ahead areas. Jobs use their own private memory for these areas and, if the job has sufficient memory, there are always areas available for read-ahead. The Excluded Command Analysis screen indicates when there are not enough read-ahead areas available because the job has exhausted local storage. A similar report occurs at the job level and is a clear indication that the local storage for the job needs to be increased. |
| Sx command ignored |
The number of Adabas |
During read-ahead processing, Adabas Fastpath stores a number of records that have been prefetched for a user, with the anticipation that the application will subsequently require these records. Inevitably, occasions arise when all the prefetched records are not required and are therefore discarded. Such records are counted as over-reads.
Adabas Fastpath uses an accelerating algorithm to decide how many records to read ahead to minimize over-reads. Consequently, it is rare to see a high level of over-reads. Because of the way the algorithm slowly accelerates the read-ahead factor, high levels of over-reads only occur on sequences with a low number of commands.
The first File Details screen for file 137 shows 668 over-reads which is 37% of read-ahead optimizations. This is high since over-reads are normally kept below 20%.
The File Details screen can be used to determine which command/descriptors are creating over-reads:
10:37:21 ***** A D A B A S FASTPATH SERVICES 7.4.1 ***** 2001-09-05
- File Details for File 137 DBID 50 SVC 240 - F12210ME
Expanded File N
On/Off Adabas Field Overrides ........Direct Access......
L1 On
L3 Off S1 S5
L9 Off S5 TA
S1 Off S5
S2 Off
Over-reads Default Adabas Field Overrides ........Read-ahead.........
L1 Off
L2 Off
82 L3 Off S5
586 L9 Off S5
S1 Off
S2 Off
S8 Off
S9 Off
Command ==>
Enter-PF1---PF2---PF3---PF4---PF5---PF6---PF7---PF8---PF9---PF10--PF11--PF12
Help Exit Excl Gen Node Menu |
This screen shows that most of the over-reads (586) result from the
L9 command with descriptor
S5. Referring back to the AFPLOOK File Details
Display, it can be seen that the L9 sequence ratio
is just over the minimum of 2. Low command-to-sequence ratios are the main
cause of high over-reads.
To reduce over-reads to less than 20%, read-ahead optimization for
the L9 command for this file can be switched off or
the descriptor override can be removed.
Although you can usually identify the command/descriptor pair causing the over-reads by finding low command-to-sequence ratios in the AFPLOOK analysis, you may occasionally need to switch off descriptors one-by-one to determine which descriptor is causing the most over-reads.
The Job Details second screen displays information about the work areas used while optimizing commands for the job. These areas are dynamically allocated and need no tuning parameters. The display is for information only:
10:55:07 ***** A D A B A S FASTPATH SERVICES 7.4.1 ***** 2001-09-05
- Job Details from Global SYSCO33 Node Id 33 - F12300MC
Job: DAEFCODE Type: COM-PLETE Number: C30694 Started Sep 5,2001 10:02
Job Information---------------------------------------------------------------
Peak Concurrency...: 2
User Areas: In Use: 2 Peak: 2 Recovered: 0
CID Areas: In Use: 0 Peak: 3
Read-ahead Memory Categories--------------------------------------------------
Size(k): 1 Peak No: 0 Usage.: 8 Failures: 0
2 0 1 0
4 0 0 0
8 0 0 0
Job Parameters----------------------------------------------------------------
Job End Stats..: N Direct Access: ON
Start Time.....: 00:00 End Time.....: 00:00
Read-ahead Optimization Control: TP Read-ahead Limit(k): 0
Command ==>
Enter-PF1---PF2---PF3---PF4---PF5---PF6---PF7---PF8---PF9---PF10--PF11--PF12
Help Exit Refr Excl Gen Node Menu |
The Job Details 2 screen shows the defined job parameters, including the read-ahead parameters. Although no explicit performance indicators exist, a general low read-ahead optimization level indicates that these parameters should be reviewed.
The read-ahead rate is accelerated automatically by algorithm: the longer the sequence continues, the more the rate increases. In this way, optimization is maximized and over-reads are avoided. Because batch jobs normally use longer sequences than online tasks, different algorithms are available: BAT for batch jobs and TP for TP monitor jobs.
You can choose either algorithm for any job. Some sites may find that the BAT algorithm works better in their particular TP systems than the default TP algorithm. The only way to determine the best option is to try out the algorithms and measure the results. This procedure applies as well when choosing the correct algorithm for such job types as TSO or TIAM.
The Read-ahead Limit parameter can be used to restrict the amount of memory allocated to read-ahead optimization for the job. This can be useful if a particular job or TP monitor is running short of local memory because of Adabas Fastpath read-ahead allocations.
