Sign-up and Commit to the Spark-GenevaERS POC

Organization of the project continues, but much progress has been made. Check out the Community Repository on GitHub, and its Governance and Technical Steering Committee Checklist to see what’s been happening.

But don’t stop there….

Get Connected

The first Technical Steering Committee (TSC) Meeting, open to all, is scheduled for Tuesday, August 11, 2020 at 8 PM US CDT, Wednesday August 12, 10 AM WAST/HK time.

Project Email List: To join the meeting or to be kept up to date on project announcements, join the GenevaERS Email List. You’ll receive an invitation as part of the calendar system. You must be on the e-mail list to join the meeting.

Slack Channel: Conversations are starting to happen on the Open Mainframe Project Slack workspace. Request access here, then when granted look for the GenevaERS Channel.

Spark POC Plans

The team continues to prepare the code for release. But that’s no impediment to our open source progress.

We’ve determined that the best way to help people understand the power of The Single Pass Optimization Engine is to contrast it–and begin to integrate it–with the better known Apache Spark.

Here’s what we plan to do work on over the next several weeks, targeting completion before the inaugural Open Mainframe Summit, September 16 – 17, 2020.

This POC was designed using these ideas for the next version of GenevaERS.

The POC will be composed of various runs of Spark on z/OS using GenevaERS components in some, and ultimately full native GenevaERS.

GenevaERS Spark POC Approach

The configurations run include the following:

  1. The initial execution to produce the required outputs will be native Spark, on an open platform.
  2. Spark on z/OS, utilizing JZOS as the IO engine. JZOS is a set of Java Utilities to access z/OS facilities. They are C++ routines having Java wrappers, allowing them to be included easily in Spark. Execution of this process will all be on z/OS.
  3. The first set of code planned for release for GenevaERS is a set of utilities that perform GenevaERS encapsulated functions, such as GenevaERS overlapped BSAM IO, and the unique GenevaERS join algorithm. As part of this POC, we’ll encapsulate these modules, written in z/OS Assembler, in Java wrappers, to allow them to be called by Spark.
  4. If time permits, we’ll switch out the standard Spark join processes for the GenevaERS utility module.
  5. The last test is native GenevaERS execution.

The results of this POC will start to contribute to this type of architectural choice framework, which will highlight where GenevaERS’s single pass optimization can enhance Apache Spark’s formidable capabilities.

Note that the POC scope will not test the efficiencies of the Map and Reduce functions directly, but will provide the basis upon which that work can be done.

Start Contributing

Open source is all about people making contributions. Expertise is needed in all types off efforts to carry of the POC.

  • Data preparation. We are working to build a data set that can be used on either z/OS or open systems to provide a fair comparison for either platform, but with enough volume to test some performance. Work here includes scripting, data analysis, and cross platform capabilities and design.
  • Spark. We want some good Spark code, that uses the power of that system, and makes the POC give real world results. Expertise needed includes writing Spark functions, data design, and turning.
  • Java JNI. Java Native Interface is the means by which one calls other language routines under Java, and thus under Spark. Assistance can be used in helping to encapsulate the GenevaERS utility function, GVBUR20 to perform fast IO for our test.
  • GenevaERS. The configuration we create we hope to be able to extract as GenevaERS VDP XML, and provide it as a download for initial installation testing. A similar goal with the sample JCL that will be provided. GenevaERS expertise in these fields is needed.
  • Documentation, Repository Work, and on and on and on. At the end of drafting this blog entry, facing the distinct chance it will be released with typos and other problems, we recognize we could use help in many more areas.

The focus for our work is this Spark-POC repository. Clone, fork, edit, commit, create pull request, repeat.

On a daily basis there is an open scrum call on these topics held at this virtual meeting link at 4:00 PM US CDT. This call is open to anyone to join.

Contributions from all are welcome.

Lucky 13: GenevaERS is an Official Open Mainframe Project project

On Thursday July 9th, GenevaERS was approved an Incubation Project under the Linux Foundation’s Open Mainframe Project.

Comments from the approving Technical Advisory Committee included (paraphrased):

  • “This project showcases the types of applications that Z is very important to.”
  • “Would like to see capabilities to cleanse data before feeding into Spark for machine learning.”
  • “Yes, intersecting with Apache Spark would be an interesting development.”
  • “This is exactly the kind of project we want to see”
  • “Most of the projects we are sponsoring are user interface related, supporting modernization. This has interesting transaction processing characteristics.”
  • “Looks like you already have a couple of people interested in participating/contributing…” The energy in the committee is “because this is a mature/sophisticated project.”

The following slides were used in the application discussion:

Additionally, the relationship of GenevaERS to Apache Spark was discussed, leveraging the content from the recent Proposed GenevaERS 5.0 Architectural Direction.

Application For Linux Foundation Project Status

The GenevaERS has applied for incubation status under the Linux Foundation’s Open Mainframe Project.

The application can be viewed here.

To graduate from Incubation Stage, or for a new project to join as an Active Stage project, a project must complete the proposal process plus:

  • Have committers from at least two organizations.
  • Have achieved and maintained a Core Infrastructure Initiative Best Practices Badge.
  • Explicitly define a project governance and committer process.
  • Have a public list of project adopters for at least the primary repo
  • Complete a code licensing scan to ensure licensing is inline with the OMP guidelines
  • Elect or appoint a project lead to represent the project on the TAC
  • Receive a supermajority vote from the TAC to move to accepted stage.

Proposed GenevaERS 5.0 Architectural Direction

The following are some initial thoughts on the next version of GenevaERS as an Open Source project might go:


Currently the only way to specify GenevaERS processes (called a GenevaERS “view”) is through the Workbench, which is a structured environment allowing specifications of column formats, values to be used in populating those columns, including the use of logic called GenevaERS Logic Text.

GenevaERS developers have known for years that in some cases a simple language would be easier to use. The structured nature of the Workbench is useful for simple views, but becomes more difficult to work with for more complex views.

Specific Proposals

In response, we propose enhancements to the front-end of GenevaERS for the following:

  • Open up a new method of specifying logic besides the Workbench, initially Java.
  • This language would be limited to a subset of all the Java functions as supported by the the extract engines.
  • The current Workbench compiler would be modified to produce a GenevaERS logic table from the Java code submitted to it.
  • Develop plug-ins for major IDE’s (Eclipse, Intellij) that highlight use of functions GenevaERS does not support in the language.
  • GenevaERS Performance Engine Processes should be able to construct a VDP (View Definition Parameter) file from a mix of input sources.


Doing this would allow:

  • Storage of GenevaERS process logic in source code management systems, enabling all the benefits of change management
  • Opening up to other languages; often extracts from GenevaERS repositories are more SQL-like because of the data normalization that happens within the repository
  • Taking in logic specified for Spark and other execution engines which conform to GenevaERS syntax, providing higher performance throughput for those processes
  • Begin to open the possibility of constructing “pass” specifications, rather than simply defined in execution scripting.
  • Perhaps creation of in-line “exit” like functionality wherein the specified logic can be executed directly (outside the Logic Table construct).

Performance Engine

The GenevaERS Performance Engine uses the Logic Table and VDP file to resolve the GenevaERS processes (GenevaERS “views”).

Specific Proposals

Proposed enhancements include:

  • Expanding today’s very efficient compiler to more functions, to support greater sets of things expressed in the languages.  This would include things like greater control over looping mechanisms, temporary variables, greater calculation potential, and execution of in-line code and called functions within a view under the GenevaERS execution engine.
  • If there is interest and capacity, we may even move in the longer term towards an alternative Java execution engine.  This would be a dynamic java engine, similar to the GenevaERS extract engine today; not a statically created for specific business functions as discussed below.

Alternative Engines

Existing customers have expressed interest in alternative execution engines for SAFR logic, as demonstrated in the Initial Open Source Work: Metadata Translation. The following are approaches that can be taken to these requests:

View to Stand-alone Java Program: It is possible to consider creating utilities which translate from GenevaERS meta data to another language, each view becoming a stand alone program, which could simply be maintained as custom processes. This would provide a migration path for very simple GenevaERS Processes to other tooling, when performance is not important.

Multi-View Execution Java Program: A set of GenevaERS views (a VDP) could be converted to a single Java program, which produces multiple outputs in a single pass of the input file, similar to what GenevaERS does today.  In other words, it is possible to look at how GenevaERS performs the one-pass architecture, isolates differing business logic constructs from each other, perform joins, etc., and write new code to do these functions.  This would also provide performance benefits from learning from the GenevaERS architecture.

Dynamic Java Program: Today’s GenevaERS compiler which produces a Logic Table could be converted to produce a Java (or other language) executable. This might add the benefit of making the processes dynamic, rather than static.  This can have benefits to changing rules and functions in the GenevaERS workbench, and some sense of consistent performance for those functions, and the potential benefit of growing community of GenevaERS developers for new functions.

Next Steps

These ideas will be discussed at an upcoming GenevaERS community call to gauge interest and resources which might be applied.

Training Module 22: using Pipes and Tokens

The slides used in the following video are shown below:

Slide 1

Welcome to the training course on IBM Scalable Architecture for Financial Reporting, or SAFR.  This is Module 22, using Pipes and Tokens

Slide 2

Upon completion of this module, you should be able to:

  • Describe uses for the SAFR piping feature
  • Read a Logic Table and Trace with piping
  • Describe uses for the SAFR token feature
  • Read a Logic Table and Trace with tokens
  • Debug piping and token views

Slide 3

This module covers three special logic table functions, the WRTK which writes a token, the RETK which reads the token and the ET function which signifies the end of a set of token views.  These functions are like the WR functions, the RENX function and the ES function for non-token views. 

Using SAFR Pipes does not involve special logic table functions.

Slide 4

This module gives an overview of the Pipe and Token features of SAFR.

SAFR views often either read data or pass data to other views, creating chains of processes, each view performing one portion of the overall process transformation

Pipes and Tokens are ways to improve the efficiency with which data is passed from one view to another

Pipes and Tokens are only useful to improve efficiency of SAFR processes; they do not perform any new data functions not available through other SAFR features.

Slide 5

A pipe is a virtual file passed in memory between two or more views. A pipe is defined as a Physical File of type Pipe in the SAFR Workbench.

Pipes save unnecessary input and output. If View 1 outputs a disk file and View 2 reads that file, then time is wasted for output from View 1 and input to View 2. This configuration would typically require two passes of SAFR, one processing view 1 and a second pass processing view 2.

If there is a pipe placed between View 1 and View 2, then the records stay in memory, and no time is wasted and both views are executed in this single pass.

The pipe consists of blocks of records in memory.

Slide 6

Similarly a token is a named memory area. The name is used for communication between two or more views.

Like a pipe, it allows passing data in memory between two or more views.

But unlike pipes which are virtual files and allow asynchronous processing, tokens operate one record at a time synchronously between the views.

Slide 7

Because Pipes simply substitute a virtual file for a physical file, views writing to a pipe are an independent thread from views reading the pipe.

Thus for pipes both threads are asynchronous

Tokens though pass data one record at a time between views.  The views writing the token and the views reading the token are in the same thread.

Thus for tokens there is only one thread, and both views run synchronously

Slide 8

An advantage of pipes is that they include parallelism, which can decrease the elapsed time needed to produce an output.  In this example,  View 2 runs in parallel to View 1. After View 1 has filled block 1 with records, View 2 can begin reading from block 1.  While View 2 is reading from block 1, View 1 is writing new records to block 2. This advantage is one form of parallelism in SAFR which improves performance. Without this, View 2 would have to wait until all of View 1 is complete.

Slide 9

Pipes can be used to multiply input records, creating multiple output records in the piped file, all of which will be read by the views reading the pipe.  This can be used to perform an allocation type process, where a value on a record is divided into multiple other records.

In this example, a single record in the input file is written by multiple views into the pipe.  Each of these records is then read by the second thread.

Slide 10

Multiple views can write tokens, but because tokens can only contain one record at a time, token reader views must be executed after each token write.  Otherwise the token would be overwritten by a subsequent token write view.

In this example, on the left a single view, View 1, writes a token, and views 2 and 3 read the token. View 4 has nothing to do with the token, reading the original input record like View 1.

On the right, both views 1 and 4 write to the token.  After each writes to the token, all views which read the token are called. 

In this way, tokens can be used to “explode” input event files, like pipes.

Slide 11

Both pipes and tokens can be used to conserve or concentrate computing capacity.  For example CPU intensive operations like look-ups, or lookup exits, can be performed one time in views that write to pipes or tokens, and the results added to the record that is passed onto to dependent views.  Thus the dependent, reading views will not have to perform these same functions.

In this example, the diagram on the left shows the same lookup is performed in three different views.  Using piping or tokens, the diagram on the right shows how this lookup may be performed once, the result of the lookup stored on the event file record and used by subsequent views, thus reducing the number of lookups performed in total.

Slide 12

A limitation of pipes is the loss of visibility to the original input record read from disk; the asynchronous execution of thread 1 and thread 2 means the records being processed in thread one is unpredictable.

There are instances where visibility by the view using the output from another view to the input record is important.  As will be discussed in a later module, the Common Key Buffering feature can at times allow for use of records related to the input record.  Also, the use of exits can be employed to preserve this type of visibility.  Token processing, because it is synchronous within the same thread, can provide this capability. 

In this example, on the top the token View 2 can still access the original input record from the source file if needed, whereas the Pipe View 2 on the bottom cannot.

Slide 13

Piping can be used in conjunction with Extract Time Summarization, because the pipe writing views process multiple records before passing the buffer to the pipe reading views.  Because tokens process one record at a time, they cannot use extract time summarization.  A token can never contain more than one record.

In the example on the left, View 1, using Extract Time Summarization, collapses four records with like sort keys from the input file before writing this single record to the Pipe Buffer.  Whereas Token processing on the right will pass each individual input record to all Token Reading views.

Note that to use Extract Time Summarization  with Piping, the pipe reading views must process Standard Extract Format records, rather than data written by a WRDT function; summarization only occurs in CT columns, which are only contained in Standard Extract Format Records. 

Slide 14

Workbench set-up for writing to, and reading from a pipe is relatively simple.  It begins with creating a pipe physical file and a view which will write to the pipe.

Begin by creating a physical file with a File Type of “Pipe”.

Create a view to write to the Pipe, and within the View Properties:

  • Select the Output Format of Flat File, Fixed-Length Fields,
  • Then select the created physical file as the output file

Slide 15

The next step is to create the LR for the Pipe Reader Views to use.

The LR used for views which read from the pipe must match exactly the column outputs of the View or Views which will write to the Pipe.  This includes data formats, positions, lengths and contents.

In this example, Column 1 of the View,  ”Order ID” is a 10 byte field, and begins at position 1.  In the Logical Record this value will be found in the Order_ID field, beginning at position 1 for 10 bytes.

Preliminary testing of the new Logical Record is suggested, making the view write to a physical file first, inspect the view output, and ensure the data matches the Pipe Logical Record, and then change the view to actually write to a pipe.  Looking at an actual output file to examine positions is easier than using the trace to detect if positions are correct as the Pipe only shows data referenced by a view, not the entire written record produced by the pipe writing view.

Slide 16

The view which reads the Pipe uses the Pipe Logical Record.  It also must read from a Logical File which contains the Pipe Physical File written to by the pipe writer view.

In this example, the view read the pipe-LR ID 38, and the Pipe LF ID 42 Logical File which contains the Physical File of a type Pipe written to by the pipe writing view.  The view itself uses all three of the fields on the Pipe Logical Record.

Slide 17

This is the logic table generated for the piping views.  The pipe logic table contains no special logic table functions.  The only connection between the ES sets is the shared physical file entity.

The first RENX – ES set is reading Logical File 12.  This input event file is on disk.  The pipe writing view, view number 73, writes extracted records to Physical File Partition 51. 

The next RENX – ES set reads Logical File 42.  This Logical File contains the same Physical File partition (51) written to by the prior view.  The Pipe Reader view is view 74.

Slide 18

This the the trace for piped views.  In the example on the left, the thread reading the event file, with a DD Name of ORDER001, begins processing input event records.  Because the pipe writer view, number 73, has no selection criteria, each record is written to the Pipe by the WRDT function.  All 12 records in this small input file are written to the Pipe.

When all 12 records have been processed, Thread 1 has reached the end of the input event file, it stops processing, and turns the pipe buffer over to Thread 2 for processing.

Thread 2, the Pipe Reader thread, then begins processing these 12 records. All 12 records are processed by the pipe reader view, view number 74.

As shown on the right, if the input event file had contained more records, enough to fill multiple pipe buffers, the trace still begins with Thread 1 processing, but after the first buffer is filled, Thread 2, the pipe reading view, begins processing.  From this point, the printed trace rows for Thread 1 and Thread 2 may be interspersed as both threads process records in parallel against differing pipe buffers.  This randomness is highlighted by each NV function against a new event file record

Slide 19

Workbench set-up for writing to, and reading from a token is very similar to the set-up for pipes. It begins with creating a physical file with a type of token, and a view which will write to the token.

Begin by creating a physical file with a File Type of “Token”.

Create a view to write to the token, and within the View Properties, select the Output Format of Flat File, Fixed-Length Fields.

Then in the proprieties for the Logical Record and Logical File to be read, select the output Logical File and created Physical File as the destination for the view

Slide 20

Like pipe usage, the next step is to create the Logical Record for the Token Reader Views to use. In our example, we have reused the same Logical Record used in the piping example.  Again, the logical record which describes the token must match the output from the view exactly, including data formats, positions, lengths and contents.

Slide 21

Again, the view which reads the Token uses this new Logical Record.  It also must read from a Logical File which contains the Token Physical File written to by the token writer view.

Note that the same Logical File must be used for both the token writer and token reader; using the same Physical File contained in two different Logical Files causes an error in processing. 

Slide 22

This is the logic table generated for the token views.

Any views which read a token are contained within an RETK – Read Next Token / ET – End of Token set at the top of the logic table.

The token writer view or views are listed below this set in the typical threads according to the files they read; the only change in these views is the WRTK logic table function, which Writes a Token shown at the bottom of the logic table in red.

Although the token reading views are listed at the top of the Logic Table, they are not executed first in the flow.  Rather, the regular threads process, and when they reach an WRTK Write Token logic table function, the views in the RETK thread are called and processed.  Thus the RETK – ET set is like a subroutine, executed at the end of the regular thread processing.

In this example, view 10462 is reading the token written by view 10461.  View 10462 is contained within the RETK – ET set at the top of the logic table.  View 10461 has a WRTK function to write the token.  After execution of this WRTK function, all views in the RETK – ET set are executed.

Slide 23

This is the Logic Table Trace for the Token Views.

Note that, unlike the Pipe Example which has multiple Event File DD Names within the Trace, because Tokens execute in one thread, the Event File DD Name is the same throughout, ORDER001 in this example.

The WRTK Write Token Function, on view 10461 in this example, which ends the standard thread processes are immediately followed by the Token Reading view or views, view 10462 in this example.  The only thing that makes these views look different from the views reading the input event file is that they are processing against a different Logical File and Logical Record, Logical File ID 2025  and Logical Record ID 38 in this example. 

Slide 24

The trace on the right has been modified to show the effect of having multiple token writing views executing at the same time. 

Note multiple executions of View 10462, the token reading view processing the same Event Record1,  from the Same Event DD Name ORDER001, after each token writing views, 10460 and 10461.

Slide 25

The GVBMR95 control report shows the execution results from running both the Pipe and Token views at the same time.

  • At the top of the report, the Input files listed include the
  • Disk File, the event file for both the pipe and token writing views
  • the Token (for the Token Reading views), and
  • the Pipe (for the Pipe Reading views)
  • The bottom of the report shows the final output disposition, showing records were written to
  • an Extract File (shared by the pipe and token reading views),
  • the Token (for the Token Writing View) and
  • the Pipe (for the Pipe Writing View)
  • The center section shows more details about each one of the Views.

Slide 26

The top of this slide shows the final run statistics about the run from the GVBMR95 control report.

Note that a common error message may be encountered if view selection for the pass is not carefully constructed, for either Pipes or Tokens.  Execution of a pass which contains only  Pipe Writers and no Pipe Readers will result in this message.  The opposite is also true, with no Pipe Writers and only Pipe Readers selected.  The same is also true for token processes.  Pairs of views must be executed for either process, matched by Logical Files containing matching Physical Files for the process to execute correctly.

Slide 27

This module described using piping and tokens. Now that you have completed this module, you should be able to:

  • Describe uses for the SAFR piping feature
  • Read a Logic Table and Trace with piping
  • Describe uses for the SAFR token feature
  • Read a Logic Table and Trace with tokens
  • Debug piping and token views

Slide 28

Additional information about SAFR is available at the web addresses shown here. This concludes Module 22, The SAFR Piping and Token Functions

Slide 29

Research on Next Generation SAFR: 2017

The following is an R&D effort to define the next generation of SAFR, composed by Kip Twitchell, in March 2017. It was not intended to be a formally released paper, but the ideas expressed herein have continued to resonate with later assessments of the tool direction.

Proposal for Next Generation SAFR Engine Architecture

Kip Twitchell

March 2017

Based upon my study the last two month, in again considering the next generation SAFR architecture, the need to separate the application architecture from the engine architecture has become clear: the same basic combustion engine can be used to build a pump, a generator, a car or a crane.  Those are applications, not engines.   In this document, I will attempt to outline a possible approach to the Next Generation SAFR engine architecture, and then give example application uses after that.

This document has been updated with the conclusions from a two day workshop at SVL in the section titled “Conclusions” after the proposal in summary.


This slide from Balancing Act is perhaps a simple introduction to the ideas in this document:

(from Balancing Act, page 107)

Key SAFR Capabilities

The SAFR architecture recognizes that building balances from transactions is critical to almost all quantitative analytical and reporting processes.  All financial and quantitative analytics begins with a balance, or position.  Creating these is the key SAFR feature.

Additionally, SAFR enables generation of temporal and analytical business events in the midst of the posting process, and posts these business events at the same time.  In other words, some business events are created from outstanding balances.  SAFR does this very effectively.

Third, the rules used to create balances and positions sometimes change.  SAFR’s scale means it can rebuild balances when rules change; for reclassification and reorganization problems, changes in required analyses, or new, unanticipated analyses.

Fourth, SAFR’s high speed join capabilities, especially date effective joins, and the unique processing model for joins on high cardinality, semi-normalized table structures creates capabilities not available in any other tool.

Last, the ability to do all these things in one pass through the data, post initial transactions, generated new transactions, reclassify some balances or create new ones, and join all the dimensions together, along with generating multiple outputs, and do so efficiently, creates a unique set of functions. 

This combined set of features is what is enable by the bottom diagram.

Key SAFR Limitations

SAFR has often been associated with posting or ETL processes, which are typically more closely aligned with the tail end of the business event capture processes.  It also has association with the report file generation processes at the beginning of the other end of the data supply chain.  It has done so without fully embracing either ends of this spectrum:  It has attempted to provide a simplified user interface for reporting processes which was not completely a language to express posting processes, and it has not integrated with SQL, the typical language of reporting processes.

The idea behind this paper is that SAFR should more fully embrace the ends of these spectrums.  However, to be practical, it does not propose attempting to build a tool to cover end-to-end data supply chain, the scope of which would be very ambitious. Covering the entire data supply chain would require building two major user interfaces, for business event capture and then reporting or analysis.   The approach below contemplates allowing most of the work on either of these interfaces to be done by other tools, and allowing them to use the SAFR engine, to reduce scope of the work involved. 

This paper assumes an understanding of how SAFR processes today to be briefer in writing.  This document is intended for an internal SAFR discussion, not a general customer engagement. 

Proposal in Summary

In summary, SAFR should be enhanced as follows:


1) The SAFR engine should be made a real-time engine, capable of real time update and query capabilities

2) It should also be configurable to execute in a batch mode if the business functions require (system limitations, posting process control limitations, etc.).

Posting Processes

3) The existing workbench should be converted or replaced with a language IDE, such as Java, Java Script, Scala, etc. The compile process for this language would only permit a subset of the total functions of those languages that fit in the SAFR processing construct.

4) When the engine is executed in “Compiled Code” mode, (either batch or real time), it would generally perform functions specified in this language, and the logic will be static.

5) Real-time mode application architecture would typically use SAFR against indexed access inputs and outputs, i.e., databases capable of delivering specific data for update, and applying updates to specific records.

Reporting Processes

6) SAFR should have a second logic interpreter in addition to the language IDE above, and accept SQL as well.  This would generally be used in an “Interpreted Logic” mode.

7) To resolve SQL, SAFR would be a “bypass” SQL interpreter.  If the logic specified in SQL can be resolved via SAFR, the SAFR engine should resolve the request. 

8) If SAFR cannot satisfy all the required elements, SAFR should pass the SQL to another SQL engine for resolution.  If the data is stored in sequential files, and SAFR is the only resolution engine available, and an error would be returned.

Aggregation and Allocation Functions

9) SAFR functionality should be enhanced to more easily perform aggregation and allocation functions. 

10) This may require embedded sort processes, when the aggregation key is not the natural sort order of the input file.

11) A more generalized framework for multiple level aggregation (subtotals) should be considered, but may not be needed in the immediate term.

12) Additionally, a generalized framework for hierarchy processes should also be contemplated.

Workshop Conclusions:

In the two day workshop I made the point that the z/OS version of SAFR does not meet the requirements of this paper; and the SAFR on Spark version did not as well.  Beyond this, the workshop concluded the following:

(1) The existing SAFR GUI and its associated meta data tables, language, and XML are all inadequate to the future problems, because the GUI is too simple for programming—thus the constant need for user exits—and it is too complex for business people—thus one customer built RMF;

(2) The database becomes a problem in the traditional architecture because there is this explosion of rows of data in the LR Buffer which then collapses down to the answer sets, but the database in the middle means it all has to get stored before it can be retrieved to be collapsed. 

(3) Platform is not an issue considered in this paper, and that the experience of both existing customers seems to demonstrate there is something unique in SAFR that other tools do not have. 

(4) We agreed that although the point about the data explosion and the database in the middle being a problem, taking on the SQL interpretation mode was a bridge too far; we should let the database do the work of satisfying the queries.  SAFR can still produce views which selects and aggregates data to be placed in the database for use, as it has historically.

(5) Stored procedures are just too simplistic for the kinds of complex logic in a posting process; so, we need to build something outside the database which can access the database for inputs and for storing outputs.

(6) There is not a good reason to require the programmer to go clear to a procedural language to specify the SAFR process; The SAFR on Spark POC SQL is a declarative language that does not put as much burden on the developer.

(7) It is possible to use in-line Java code within the SAFR on Spark prototype when a procedural language is required.  This could eliminate simple user exits.

(8) Spark is a good framework for a distributed work load; z/OS manages the parallel processes very effectively, but requires a big server to do so.  Using another language on another platform would require building that distribution network.  That is expensive.

(9) The real time capabilities would simply require creation of a daemon that would then execute the SAFR on Spark processes whenever needed.  This is nearly a trivial problem; performance and response time notwithstanding.

(10) We do not have enough performance characteristics to say if the Spark platform and the prototype will scale for any significant needs.

Real-time Engine and IO Models

The following table outlines likely processing scenarios:

Real-time ModeBatch Mode
Input TypesDatabase Messaging QueueSequential files or Database or Messaging Queue
Output TypesDatabase or Messaging Queue or Sequential FilesSequential Files

Sequential files in this table included native Hadoop structures.  Output from batch mode files could be loaded into databases or messaging queues beyond SAFR processing, as is typically done today with SAFR.

Recommendation:  SAFR today recognizes different IO routines, particularly on input.  It has an IO routine for sequential files, DB2 via SQL, VSAM, DB2 via VSAM. It has had prototype IMS and IDMS drivers.  This same construct can be carried forward into the new version, continuing to keep separate the logic specification from the underlying data structures, both inbound and outbound.  New data structure accesses can be developed over time.

Enhancement:  An additional processing approach could be possible but less likely except in very high volume situations: Real-time mode against input sequential files.  In this configuration, SAFR could continually scan the event files.  Each new request would jump into the ongoing search, noting the input record count of where the search begins.  It would continue to the end of the file, and through the loop back to the start of the file, continuing until it reaches the starting record count again, and then end processing for that request.  This would only be useful if very high volumes of tables scans and very high transaction volumes made the latency in response time acceptable.

Impediments:  The challenges to a real-time SAFR engine is all the code generation interpretation processes in the performance engine.  Making these functions very fast from their current form would be difficult.  Using an IDE for logic specification has a much better chance of making the performance possible.  Performance for the interpreted mode would require significant speed in interpretation and generation of machine code from the SQL logic.

Compiled Code Mode

Background:  Major applications have been built using SAFR.  The initial user interface for SAFR (then called Geneva, or 25 years ago) was intended to be close to an end user tool for business people creating reports.  The Workbench (created starting in 1999) moved more and more elements away from business users, and is now wholly used only by IT professionals. 

Additionally, because of the increasingly complex nature of the applications, additional features to manage SAFR views as source code have been developed in the workbench.  These include environment management, and attempts at source code control system integration.  These efforts are adequate for the time, but not highly competitive with other tools. 

SAFR’s internal “language” was not designed to be a language, but rather, only provide logic for certain parts of a view.  The metadata constructs provide structures, linkage to tables or files, partitioning, join logic.  The view columns provide structure for the fields or logic to be output.  The SAFR language, called Logic Text, only specifies logic within these limited column and general selection logic.

Recommendation:  Given the nature of the tool, it is recommended SAFR adopt a language for logic specification.  I would not recommend SQL as it is not a procedural language, and will inhibit new functionality over time.

The fundamental unit for compile would be a Pass or Execution, including the following components:

  • A Header structure, which defines the meta data available to all subprograms or “method”.  It would contain:
    • The Event and lookup LR structures
    • The common key definition
    • Join paths
    • Logical Files and Partitions
    • Available functions (exits)
  • A View Grouping structure,
    • Defining piping and token groups of views.
    • Tokens and similar (new) functions could potentially be define as In-line (or view) function calls (like tokens) which can be used by other views.
  • Each view would be a method or subprogram that will be called and passed each event file record

The final completed Execution code could be compiled, and the resulting executable managed like any other executable; the source code could be stored in any standard source code control library.  All IT control and security procedures could follow standard processes.  This recognizes the static nature of most SAFR application development processes.

Although this approach offloads some portion of the user interface, language definition, and source code management issues, the next gen SAFR code base would have to include compilers for each platform supported.  Additional SAFR executables would be IO modules, and other services like join, common key buffer management, etc. for each platform supported.

Impediments:  A standard IDE and language will likely not include the Header and Grouping structure constructs required.  Can a language and IDE be enhanced with the needed constructs? 

Interpreted Logic Mode

Background:  Certain SAFR processes have been dynamic over the years, in various degrees.  At no time, however, has the entire LR and meta data structure definition been dynamic. In the early days, the entire view was dynamic in that there were fairly simple on-line edits, but some views were disabled during the production batch run. In more recent years fewer pieces have been dynamic.  Reference data values have always been variable, and often been used to vary function results, in the last two decades always from externally maintained systems like ERP systems.  Most recently, new dynamic aspects have been bolted on through new interfaces to develop business rules facilities. 

The SAFR on Spark prototype application demonstrated the capability to turn SAFR logic (a VDP, or the VDP in XML format) into SQL statements quite readily.  It used a CREATE statement to specify the LR structure to be used.  It had to create a new language construct, called MERGE to represent the common key buffering construct.  SQL tends to be the language of choice for most dynamic processes.  SQL also tends to be used heavily in reporting applications of various kinds. 

Recommendation:  Develop a dynamic compile process to turn SAFR specific SQL into SAFR execution code.   This will allow for other reporting tools to gain the benefits of SAFR processes in reporting.  If the data to be accessed by SAFR is in a relational database (either the only location or replicated there), SAFR could pass off the SQL to another resolution engine if non-SAFR SQL constructs are used. 

Impediments:  It is not clear how the MERGE parameters would be passed to the SAFR SQL Interpreter.  If sequential files are used for SAFR, would a set of dummy tables in a relational system provide a catalogue structure sufficient for the reporting tool to generate the appropriate reporting requirements?  Also, can the cost of creating two compilers be avoided?

Enhancement:  CUSTOMER’s EMQ provides a worked example of a reporting application which resolves SQL queries more efficiently that most reporting tool.  It uses SQL SELECT statements for the general selection criteria, and the database functions for and sort (ORDER BY), and some aggregation (GROUP BY) functions.  When considering SAFR working against an indexed access structure, this construct may be similar.  More consideration should perhaps be given.

Compile Code and Interpreted Logic Intersection

Background:  The newer dynamic, rules based processes, and the older, more static Workbench developed processes cannot be combined into a single process very easily because of duplicated metadata.  This arbitrary division of functions to be performed in Compile Code mode verses Interpretive Mode is not desirable; there may be instances where these two functions can be blended into a single execution.

Recommendation: One way of managing this would be to provide a stub method which allows for calling Interpreted processes under the Compiled object.  The IDE for development of the Interpreted Mode logic would be responsible for ensuring conformance with any of the information in the header or grouping structures.

Aggregation and Allocation Processes

Background: Today’s SAFR engine’s aggregation processes are isolated from the main extract engine, GVBMR95, in a separate program, GVBMR88.  Between these programs is a sort step, using a standard sort utility.  This architecture can complicate the actual posting processes requiring custom logic to table, sort (in a sense, through hash tables) detect changed keys, and aggregate last records before processing new records. 

Additionally, today’s SAFR does not natively perform allocation processes, which require dividing transactions or balances by some set number of additional records or attributes, generating new records.  Also, SAFR has limited variable capabilities today. 

Recommendation:   Enhancing SAFR with these capabilities in designing the new engine is not considered an expensive development; it simply should be considered and developed from the ground up.

Next Steps

Feedback should be sought on this proposal from multiple team members, adjustments to it documented, and rough order of estimates of build cost developed.