Thursday, February 01, 2018

Db2 Application Development for Performance: Be Early and Be Informed

The title of this blog post is also the title of one of my two IDUG Db2 Tech Conference presentations in Philadelphia this year. I chose this topic because I am in the process of writing a book on Db2 performance from a developer perspective.

There are a lot of performance-focused presentations but most of them are from the perspective of monitoring and tuning. In other words, they are about looking for and fixing problems in code that already exists. Now there is nothing wrong with this. In fact, it is necessary. Even if everything is written correctly from the get-go (right, sure it is) over time things will change and performance will decline. So we need tools to ferret out what is going wrong and help us to fix it. This is mostly the domain of the DBA to do.

From the programmer's point-of-view, though, it seems that things are lacking. I frequently see SQL and program code that seems to have been written by someone with no understanding of relational basics. This has to change.

That is the reason for my IDUG presentation... and also for the book. The idea is to give advice and direction to Db2 application developers and programmers on writing efficient, well-performing programs. If you write code and access data in a Db2 database, then this book and presentation should be helpful to you.

The material is written for all Db2 professionals. It will be useful for users of Db2 for Linux, Unix and Windows as well as for users of Db2 for z/OS. When there are pertinent differences between the two I point it out. Also, much of the material will apply to any relational database system, not just Db2; nevertheless, the book is written and intended for Db2 users.

I don't try to teach basic programming skills, other than, of course, some guidance on SQL. And even that is not basic. I assume you can figure out a basic Select, Insert, Update or Delete... and even some more complex stuff like joins. The focus of the book is on programming, coding and developing applications.

If you are a DBA, most of the material will still be of interest to you because DBAs are responsible for overall Db2 performance. Therefore, it makes sense to understand the programming aspect of performance. That said, I will not be covering DBA and system administration level performance. But if you think the material is worthy, knowing it is there to recommend to your developers (new and old) can be worthwhile!

The book and presentation focus on guidance regarding application development procedures, techniques, and philosophies. The goal is to educate developers on how to write good application code that lends itself to optimal performance. Of course, this does not mean that every program you write will perform at top speed. But it should enable you to write code that does not require remedial after-the-fact modifications by performance analysts. If you follow the guidelines I outline in this book and presentation, I can say with confidence that your DBAs and performance analysts will love you!

So if you are going to IDUG in Philadelphia this year, be sure to attend my presentation. Be sure to say "Howdy!" and that this blog post guided you to the presentation... and also, keep an eye on my web site, Mullins Consulting, Inc., for information about the book when it gets published, hopefully later in 2018.

Monday, December 25, 2017

Seasons Greetings 2017

It is that time of year again... A time to reflect on the year gone by and to enjoy time with our friends, family and loved ones.  A time to put down the work that consumes us most of the year and to celebrate and enjoy... to remember past years and to consider the upcoming year.  

No matter what holiday tradition you celebrate, I wish you an enjoyable holiday season. Seasons greeting to one and all... and I'll see you next year here on the Db2 Portal blog!

Monday, December 18, 2017

The Db2 12 for z/OS Blog Series - Part 20: Fast Insert: An Alternative INSERT Algorithm

Db2 12 offers many performance improvements that can help you to speed up your applications. The Fast Insert algorithm, also called Insert Algorithm 2, is one such improvement, but you have to understand what it is and when it can be helpful.

The general idea behind this alternative Insert algorithm is to improve performance of Insert-intensive workloads for unclustered data. This is a common requirement for applications that require an audit table or a journal of activities. New data that is inserted is appended to the end of the table with no concern for clustering. A frequent issue with such applications is when the workload is so high that rows cannot be inserted rapidly enough thereby increasing the elapsed time.

Most of these types of applications design the journal/audit table using partitioned table spaces with the MEMBER CLUSTER and APPEND attributes. This design will direct Db2 to insert all new rows at the end of the partition. Insert performance should improve because the space search algorithm can be bypassed for the most part. Nevertheless, some Db2 applications still experienced performance issues even when using this approach.

This bring us to the Db2 12 Fast Insert algorithm. This new algorithm uses an in-memory structure called an insert pipe to speed things up. The insert pipe maintains a list of pages that are available for this member to use for Insert processing for the page set partition. Each member that opens a page set gets an Insert pipe for that member to use. A system agent fills up the pipe asynchronously, making sure that pages are always available for the threads to use for inserting rows.

There is more to the process, but that is the high-level intro to how it work. Now the question is: when will the new algorithm be used?

The Fast Insert algorithm only works with Universal table spaces with the MEMBER CLUSTER option; APPEND is not required. The new algorithm is the default algorithm for this type of table space.

Settings and options are available to control use of the new algorithm. To set usage of the algorithm at a system-wide level, use the DEFAULT_INSERT_ALGORITHM subsystem parameter (DSNZPARM). There are three options:
  • 0 indicates that the basic Insert algorithm is to be used no matter what (Insert algorithm 2 is disabled for the subsystem); 
  • 1 indicates that the basic insert algorithm is used unless insert algorithm 2 was specified at the table space level; 
  • 2 indicates that Insert algorithm 2 is used unless insert algorithm 1 was specified at the table space level.

Which brings us to the DDL options for controlling the insert algorithm at the individual table space level. To do so, use the INSERT ALGORITHM option on the CREATE TABLESPACE or ALTER TABLESPACE statement. Again, there are three options: 

  • 0 indicates that the Insert algorithm to be used is as specified by the DEFAULT_INSERT_ALGORITHM subsystem parameter at the time a row is inserted; 
  • 1 indicates that the basic Insert algorithm is to be used; and 
  • 2 indicates that the Fast Insert algorithm is to be used.

Summing Up

The impact of using the new Fast Insert algorithm will depend on various factors, including whether the table has indexes or not and the specific makeup of your workload. If the workload has lock/latch contentions (on the space map pages and data pages) then the new Insert algorithm will probably be beneficial.

Thursday, November 23, 2017

Happy Thanksgiving 2017

Today, November 23rd, in the United States of America, we celebrate Thanksgiving by gathering together with our loved ones and giving thanks for what we have. 

Typically, this involves celebrations with food, traditionally a big turkey dinner with stuffing, mashed potatoes and gravy, as we watch a parade and football games. I plan to follow this tradition to the letter this year and I wish you the same!

And with the general notion of giving thanks on this fine day, I want to also pause and thank each and every one of you for reading my blog, whether just today, or all year long. Hopefully you'll keep coming back and we can keep discussing all things Db2 here.

But for today, wherever you may be, I am thankful for you and I wish you a very happy day!

Oh... and try not to get trampled on Friday if you're going out shopping for bargains!

Monday, November 13, 2017

The Db2 12 for z/OS Blog Series - Part 19: Profile Monitoring Improvements

The ability to monitor Db2 using profile tables is a newer, though by no means brand new capability for Db2 DBAs. You can use profile tables to monitor and control various aspects of Db2 performance such as remote connections and certain DSNZPARMs.

But this blog post is not intended to describe what profile monitoring is, but to discuss the new capabilities added in Db2 12 to enhance profile monitoring.

There are four new enhancements offered by Db2 12 for the use of system profiles.

The first enhancement is the ability to automatically start profiles when you start up a Db2 subsystem. This can be accomplished using a new subsystem parameter called PROFILE_AUTOSTART. Setting the parameter to YES causes Db2 to automatically execute START PROFILE command processing. The default is NO, which means that Db2 will not initiate START PROFILE when the subsystem starts up.

The second improvement is the addition of support for global variables. As of Db2 12 you can specify the following global variables as a KEYWORDS column value in the SYSIBM.DSN_PROFILE_ATTRIBUTES table:

If a profile filter matches a connection, Db2 will automatically apply the built-in global variable 
value to the Db2 process of that connection when the connection is initially established, and when a connection is reused.

Wildcarding support is the third enhancement for profiles in Db2 12. One row for each profile is contained in the SYSIBM.DSN_PROFILE_TABLE. Each column in the table informs Db2 which connection to monitor. Without wildcarding, handling various connections required multiple rows to be defined in the table. But with Db2 12, you can have one row representing more than one connection. Wildcarding is available for AUTHID (authorization IDs), LOCATION (IP addresses of monitored connections), and PRDID (product specific identifier, for example DSN for Db2).

The fourth and final enhancement is for managing idle threads. The MONITOR IDLE THREADS column in the SYSIBM.DSN_PROFILE_ATTRIBUTES table directs DB2 to monitor (for an approximate amount of time) an active server thread’s idle time. The ATTRIBUTE1 column, which is used to specify the type of messages and level of detail of messages issued for monitored threads, has been enhanced to allow the following values: 

Note: This particular change to idle threads
for EXCEPTION_ROLLBACK was made available
in Db2 11 after general availability, and will be
available on a Db2 12 system after new function
is activated.

For more details on any of these capabilities, or indeed, on profile monitoring in general, refer to the IBM Db2 12 for z/OS Managing Performance manual, SC27-8857.