Day 660 - "Go" Decision For Half Data Migration

This situation and the pressure people is getting to make things fast it's totally unbelievable. This has lead management into splitting the data migration by the next two weekends. Thus, the next weekend it will migrate customers, insurance claims and invoices. But only if they are not involved in some kind of accident, otherwise it's migration will happen in the other weekend.

This will allow the integration test teams to perform preliminary validations and I also believe it will be great because I know that the entire data migration does not integrate into GIS during a single weekend with the current AS/400 system configuration.

Meanwhile, the software patches continues to give us problems.

297 - Data Loading Files Performance Problem

The migration scope has increased.
New source data has come into play and thus new mapping specifications have been made.

This has resulted in a really big data loading performance problem.
Our data transformation procedure is still fast, but the creation of the GIS loader files from the transformed data is starting to give us some headaches.

The GIS data loader has a flat file structure that is very verbose, the loading of each single value of each record is done through a 300 character text file line.
This means that a single database table row is transformed into, something like, the same number of text lines as the number of columns the row has. Plus the file and records header and footer structure.

As an example, house insurances data transformation is performed in around 4 hours, sequential time, and it generates about 184 millions of records. This is all performed in the transformation server.
These records are then exported into the GIS data loader file format from the Windows directly into the AS/400. This procedure is now taking much time, over 6 hours, sequentially, in the best case scenario.
This is obviously too much time, so we are exploring several hypotheses, ranging from creating a parallel file write process; writing the files locally, with and without compression, and transfer them via FTP to AS/400; cluster indexes with full coverage maintained in a different disk; splitting the database schemes across several disks.
Some of these techniques can, and will, be combined.

We have to tune the process from our side since there is not much the GIS or the AS/400 can do when it comes to massive data load tuning.
We are facing a lot of hard work in these next days.

Day 240 - Cleanup Procedure Status

The AS/400 clean up procedure and tuning procedures has resulted in a better performance.
The performance increase was not substantial, but since the domain includes millions of records, any minor improvement results in an overall improvement.

Day 232 - Loading Perfomance Problem

Finally we were able to start the load of the last weekend data into GIS for testing purposes.
The entities were loaded with less than five thousand rejects on a near one million records.
It is not a bad ratio, but I was expecting a lower rejection rate by now.

The loading of a subset of the car insurance claims, around 450000 records, as also started.
Most of car insurance claims are loading without errors, but a critical problem has raised.
GIS is loading around 2100 records per hour.
This means that 50000 records will require 24 hours to load, and the full set will require an impossible 10 days to be accomplished.
This is sequential time, but even if we use 6 CPUs at a time it will still require more than one day and a half to accomplish this task. Plus, the AS/400 were GIS is running is unavailable 3 hours every night for maintenance procedures.

This is more than a technical challenge, it is a critical situation that will require the involvement of management in the process of finding a solution.

Day 83 - Data Loaded into DB2 on Windows

We have finally received the database data from OS/390 in a format that we were able to load it without trouble in DB2 on Windows.

Some of the weird problems we got on AS/400 will no longer bother us but now, we have to rewrite some scripts because DB2 in Windows is different from DB2 on AS/400.

One of the problems was that I an insert into script that loads the postal code with almost 300000 records. I had to loaded it through the DB2 command line tool. Here's how one can do it:

1. Open the DB2 Command Line Processor tool. Don' t just open a regular shell since it will not work, some environment settings are required.
2. In the shell, execute the following commands:
   

db2 connect to {database} user {username} using {password}
db2 -vf {path/script.sql} -t > output.txt

Don't forget to have a tail and grep installed, if you wish to track the evolution of the script execution and check for operation success.

The entire process took about 6 hours, this includes taking the data from OS/390, moving it to Windows, load it into DB2 in Windows and preparing the staging area with look up and convert tables. Way to much time. This process will definitely be optimized.

We've also found out one of the most idiotic functionality in software I've ever seen.
After installing the database client on the local machines, and after configuring the server access, the Command Editor only allows one to open files from the database server computer. Yes, I mean it is impossible to open any local file using the Command Editor. And no, that is not configurable!

Day 59 - Perfomance Tests on Intel

We've finished the performance tests on the Intel server.
It's an 8 CPU machine with 8 GB of RAM running Windows Server 2003 64 bits. I was hopping to get it running Linux, but that seem not to be possible.

If you recall, our transformation engine it CPU bound and each transformation engine runs on a single CPU, therefor a direct comparison is easy to perform.

The server is powerful enough to hold the transformation engine and the database itself. Therefor, we made several tests combining the servers.
One test case used the AS/400 as the database server and the Intel machine as the transformation engine. On another test case, both the database and the transformation engine were on the Intel machine.
For easiness, the database on the Intel machine is a, real, DB2.

Amazingly, or not, the Intel CPU was a lot faster than the AS/400 CPU. One single Intel CPU runs our allows us to migrate faster than any combination we've tried using the AS/400.

Here's a full graphic showing the results on both systems. It's easy to see that the Intel machine scales very well when we use several processes in parallel.

Day 46 - Performance Tests

I know it's Saturday, but a lot of data migrations work is done during the weekends. Specially during the last weeks before the end of the project.

But this Saturday, we were focused on performance tests. We had the AS/400 entirely for us for about 3 hours.
We executed performance tests for the transformation rules engine and for the GIS data loader.

As excepted, the performance increased both with a single process and with multiple processes. for both systems.
In the transformation engine we've executed up to 6 parallel migrations, but the maximum gain was below that, the 4th or 5th, depending on the transformation features, more I/O bound or more CPU bound.

There was not time to perform one interesting test though, running the transformation engine at the same time as the GIS data loader.

Nevertheless, the outcome was what we expected: to have a dedicated Intel machine running Linux for the exclusive usage of the Data Fusion. This will allow GIS data loader to have more resources, since we're not competing.

Here's some nice charts from the tests performed.

Day 41 - DB2 Problems, Part 2

As expected, and as I've logged on Day 28, DB2 on AS/400 is giving us a bad time.

DB2 on AS/400 is not the "real" DB2. Is something awkward, that usually behaves like DB2 and sometimes does not behave at all.

We've used DB2 on other data migrations and we've only had problems on AS/400 version.

To get a glimpse, here's some problems we're having:

• trying to create a view is impossible since DB2 returns a generic error saying it is not possible to complete the command.
• trying to select specific fields from a view is impossible since DB2 returns an erro saying it does not find such fields.

One could think I may have extremly complex queries, well, judge for yourself (real column and table names have been changed).

Running the following query works fine:

select substr(ZIP1, 1, 2) ZIPinit, city,
       (ZIP1 || '-' || ZIP2) as zip_code
from (
    select ZIP1, ZIP2, city, 
    row_number() over (partition by substr(ZIP1, 1, 2), city 
    order by ZIP1, ZIP2, city) as ZIP_ORDER
from LIBWRK.ZIP_CODES
) as ZIPs

but creating a view over it is impossible:

create view LIBWRK.ZIP_APROXIMADO as
  select  substr(ZIP1, 1, 2) ZIPinit, city,
         (ZIP1 || '-' || ZIP2) as zip_code
  from (
    select ZIP1, ZIP2, city,
      row_number() over (partition by substr(ZIP1, 1, 2), city
      order by ZIP1, ZIP2, city) as ZIP_ORDER
      from LIBWRK.ZIP_CODES
  ) as ZIPs
  where ZIP_ORDER = 1

Also, it is not possible to perform all queries over this view:

create view LIBWRK.ZIP_CONVERTION as (
  select ZIP.ZIP1 as ZIP1, ZIP.ZIP2 as ZIP2, 
    ZIP.cityDescricao as cityDescription,
    ZIP1 || '-' || ZIP2 as ZIP_CODE,
    substr(ZIP1, 1, 2) as ZIPinit,
    case 
       when ZIP.ZIP2 = '000' then 
          ZIP1 || '-' 
          || (select min(x.ZIP2)
              from LIBWRK.ZIP_CODES as x 
              where x.ZIP2 <> '000' and x.ZIP1 = ZIP.ZIP1)
       else ZIP1 || '-' || ZIP2
    end as zip_code,
    ZIP.NormalizedCity as NormalizedCity,
    row_number() over (
             partition by substr(ZIP1, 1, 2), NormalizedCity
    order by ZIP1, ZIP2, NormalizedCity) as ZIP_ORDER
    from LIBWRK.ZIP_CODES as ZIP
);

select * from LIBWRK.ZIP_CONVERTION

works fine, but

select ZIP_CODE from LIBWRK.ZIP_CONVERTION

does not work at all.

But DB2 on AS/400 has a lot more problems. We've still not found them here but have runned into this in previous data migrations:

• query exploded when running a query: sometimes the query runs fine, some times it explodes, and yes, "query exploded" is the exact term that DB2 writes on its log;
• problems with summary querys: the optimizer was unnable to optimize de query and it tooks so much time it it becomes impossible to run such queries.

The "pragmatic" solution for all these problems has been creating tables populated with the queries result. In cases like the summaries, it was necessary to build sereveal summary tables with small subsets of the data and then create a view, if possible, or a table that joinned all the pre-calculated data.

Day 34 - Denial Phase

The meeting with the project sponsor went almost as expected.
They loved the idea of having a separate server to execute the transformation rules, but the AS/400 administrators are in denial.
AS/400 administrators don't want to believe that AS/400 is not that as fas as they all brag about. If I had spent as much money as an AS/400 costs, I would be denying those facts too... It's just too painful to hear that in certain tasks, the AS/400 can be beaten by a much cheaper 8 CPU Intel machine running Linux...

By the way, the AS/400 were we're testing, the development environment, has the following features:

• iSeries 570
• 1 CPU Power 5 at 1.65GHz
• 3000 CPW
• 10GB RAM DDR1
• Internal HD: 1.6TB

The AS/400 administrators now want us to perform tests on the qualification machine:

• System i Power 6 570
• 1 CPU a 4.7GHz (maximum of 4 processors)
• 5000 CPW (maximum of 20000 CPWs)
• 20GB RAM DDR2 400MHz (maximum of 90GB)
• External HD: 2.4TB

I bet the test will turn out the same results, since the transformation engine is CPU bound, but nevertheless, we'll do the test during a weekend where there's no activity on the machine.

./M6

Day 30 - Performace Tests

There's a set of performance tests to do before choosing the final migration infrastructure.

There's tests over the initial data extraction from OS/390 and loading into AS/400.

There's tests over the data transformation rules.

There's tests over the final data loading into GIS.

From this set of tests, only ours, the data transformation rules, have been performed so far.

As expected, AS/400 is not a fast machine to run Data Fusion transformation engine. We already knew this, but AS/400 system administrators always say we're wrong and that they have a fantastic and fast machine. Their pride has been hurt...

As an example, running entirely in AS/400, 400000 source records have migrated at ~800 records per secord. Note that this was a local data migration execution, the source and target databases are on the same machine as the transformation engine.

But running the same transformation engine on my, not very fast, desktop computer with 1GB of RAM, the same 400000 records migrated at ~1400 records per second. Note that on this test the source data was retrieved from AS/400, transformed on my desktop compute, and sent back to AS/400, i.e. the data had to travel over a common 100 Mbit network.

The protocol loading interface was also subject to tests and it's CPU bounded, We have one single processor, of a 4 processor machine, and it consumes 100% of that CPU. When running on AS/400, it perfomed ~12000 records per second. Running on the desktop it does only ~8000 records per second.

The scalability tests on AS/400, with a single CPU for us, proved that a linear execution is faster than a two parallel executions. We haven't had the chance to test parallel execution with more CPUs.

We've been discussing what was the best approach allow us to scale and one thing is for sure, we need to run the transformation engine outside the AS/400, because on the D day we have to share, as it's happening now, the machine with others, specifically with GIS data loading.

Probably we'll have a 6 or 8 CPU machine running Linux and connected to AS/400 over a 1 Gbit network. This will allow us to run multiple transformation engines and free AS/400 for the other data migration processes.

Day 28 - DB2 Problems

We've already used AS/400 and DB2 on previous migrations.

Our experince tell us that they're not very good when it comes use it as the main data migration machine.

There's a set of problems on DB2 that we were afraid to find again on this AS/400 version. This is the latest AS/400 version and, in the words of the system administrator, currently it may be the fastest AS/400 in Portugal.

Our pessimistic expectations proved to be right. DB2 continues to have problems, with seem to be bugs on its engine.

For instance, executing an query works fine, but creating a view over that same quety is impossible because DB2 crashes with an internal error!

I had to create a new table with data from the resulting query.

I stil have not found the "query exploded" problem, which seem related with DB2 optimizer that crashes when trying to optimize the query.

Our confidance on DB2 on AS/400 is a bit low right now, since we proved to be right when we said that we were expecting problems from it.

Day 24 - New Mapping Rules

There was a new meeting for mapping rule definition. Most of the transformation rules are closed, there's just some tweaks related with foreign addresses and some internal codes.

Next week there will be more full migration cycles with fresh data, since we've discovered that the current test data was not correct.

Meanwhile, performance tests are being performed on AS/400 and we were able to convince the system administratos to reconfigure the anti-virus not to read Java .jar files, allowing a better performance on our desktops.

And yesterday, the desktops were automatically updated and rebooted. This policy has also been removed from our desktops, since it is far too dangerous to perform such operations when a data migration is being controled from our desktops.

Day 17 - First Try

The protocol loading interface is ready, it's cool but I did not code it.

And I have the first mappings ready to roll.

We've done a full cycle, with about 85% of the transformation rules implemented, and we got less problems than I expected for the first try. 

Foreign zip codes were wrong, which we were expecting since this mapping specification is still under discussion.

People names were wrongly specified, since the surname was being lost.

The were some empty honoricif titles, since they are also still under specification.

A couple of bytes were on the wrong place on the protocol file.

We've tried it with one hundred records from the source, which resulted in, aproximatly, 2600 records in the protocol file.

The whole process was executed in our desktop computers, since the AS/400 is still not ready for us, the Java 1.5 is missing. Next week I'm expecting to execute the whole process in the AS/400.

Day 3 - Target specification

The target specification arrived today.
It's an hierarchical structure that will result in a flat file where each line will represent a specific piace of information, identified by a code.
We've started to work on it, but since the specification is more like a meta-model, we had to make some adjustments to comply with such a dynamic structure.
All problems have been solved, they were actually simple decisions that the source team helped to make.
The DB2 model has been created and our scripts turned it into mapping sheets, that will be used in the functional mapping performed by the source and the target teams.

A piece of the source data has become available. It has been extracted from the OS/390 in EBCDIC and in text formats in order to allow some AS/400 loading tests.
The data looks correct, but each row has been loaded as a single record, the copybook rules have not been applied and therefor, it will have to be loaded again. Nevertheless, the fact that the information looks correct is a good enough as the result of the first try.

The first workstation also arrived today, allowing us to install and configure some of the software required for the data migration, and allowing us to access to the development AS/400 machine.

It will not depend on us, but if next week things work out the way we've planed, we'll be able to recover the one day delay.

Day 1 - Kick off

The project has been officially kicked off.

There was the traditional meeting where all teams where present and an overview of the project was presented. All teams involved have previous data migration experience, which is good.

The project should be completed around mid October, but I felt that management confidence to comply with the date is not very high. This is not a bad signal as one might think, since management showed that it is aware of some real difficulties. Those difficulties may become technical challenges and may delay the project.
Nevertheless, I do believe that the original due date can be achieved.

There will be two phases. The first phase, the current one, will use a small entity subset in order to test the performance of the ETL and to tune the process.
The second phase will be the continuity of the previous phase, but now including all applications and entities.


The project itself consists of migrating the current insurance application, running on OS/390 mainframe, into the new system, the GIS Non-Life, running on AS/400, all using DB2, SAM and VSAM.
There are some other satellite applications running on SQL Server 2005 on Windows.
The ETL tool will be Data Fusion and it will run on AS/400 under the Qshell.

The ETL process will be implemented in a slightly different way from the usual scenario. Usually the data is extracted from the source database and loaded into the target database. This is not the case in this project.
Eventually, the data will be loaded into the final target database, but there's an intermediate step. The ETL process performed by Data Fusion will deliver the data on flat files that will be later consumed by the GIS loader. The GIS loader will validate and load the data into the final database.

Since our (ETL team) experience tells us that performance on AS/400 may be a problem during the ETL process, there was some discussion over the performance tests to be performed during phase one.

If the performance becomes an issue, the data migration may have to be incremental, instead of one-shot. This will be a problem because the source team cannot identify which data has been changed and needs to be migrated again to refresh the target system. One challenge that has been given to us, is to think in a fast way to identify if a record has changed and needs to be migrated again, just in case of a B plan is required.