Hi all, and welcome to my new blog!
All this week, I am busy at the International DB2 User's Group conference here in Athens, Greece. The two topics I'm presenting are "Native SQL stored procedures in DB2 9 for z/OS" and "Java stored procedures".
I was very glad to get the chance to get across this key point in yesterday's presentation on native SQL stored procedures -- yes, when they execute they are eligible for redirect to the zIIP processor, but only when they are invoked from a remote client, and then at the same percentage as other DDF work. Lots of presentations lately have stated this a bit more broadly, leaving people with the wrong impression. So, let's be clear - when a remote thread comes into DB2, it executes on an enclave SRB, and DB2 dials the zIIP redirect to a certain percentage. A DB2 9 native SQL procedure executes on the invoking execution block and not on a WLM-SPAS TCB - that's one of their big advantages. Thus, when a native SQL stored procedure is invoked from a remote client over a TCP/IP connection, it runs on the enclave SRB and thus picks up that same DDF zIIP redirect percentage. On the other hand, when a native SQL stored procedure is invoked locally on z/OS, it is executed on the TCB that PC'd in from CICS or batch, and that is not eligible for zIIP redirect.
Lots of other good stuff going on here - tomorrow is a DB2 for z/OS Special Interest Group as well as our IBM query panel.[Read More]
Last week in Athens I attended a presentation by Julian Stuhler of Triton Consulting
. I was of course aware of the support, but more from a DB2 internals point of view. It was great to get an external perspective on it from someone who has been working with it.
The key information is that spatial data can be points, lines, or polygons (including multi-part polygons). If you think about it, this is really powerful. One example that Julian used is that an address is a point, and a flood zone is a polygon. So now you can ask "is the house in a flood zone", which is "is this point inside the polygon"? Cool stuff! I can really imagine how this could be used by some situational applications to use data in DB2 alongside other data.
Complete documentation on the spatial features can be found in this book
I often get asked about an architecture where a stored procedure is used for a single SQL statement. This is one of the most common errors in designs using stored procedures. It's always best to amortize the CALL overhead by including several SQL statements and even some business logic in every stored procedure. But folks that come to DB2 for z/OS from other DBMS's still do this.
A new twist on this is with our DB2 9 for z/OS support for native SQL procedures. Folks ask me, well now is it OK to have a single SQL statement in a native SQL procedure?
The thing is, the native SQL procedure is still a package for DB2 to load or at least switch to. So there really still is overhead, not to mention the network time to get over to DB2, as most apps will have to invoke several stored procedures to accomplish their logic. And guess what? Those applications typically invoke the same stored procedures in the same order with the same application logic in between. So... why not make that whole set a single stored procedure?
Bottom line: even with native SQL procedures executing in the DB2 engine rather than the WLM-managed address space, I still recommend an architecture with multiple SQL statements and some business logic rather than a single SQL statement per procedure.
If you do have an architecture with single SQL statements in stored procedures, that's not to say it won't work - it will work, just not as efficiently as it would if you follow the above advice. And even if the single-threaded app performs OK, it won't scale as well as having more SQL statements and logic in the stored procedure, due to longer-held locks over network transmissions as well as some database engine serialization required for external SQL procedures (that part at least is gone in DB2 9).
SQL procedures are absolutely strategic and I do recommend an architecture based on them. The advice I've given other customers is to determine the ones that are invoked most often, analyze the patterns of when they are invoked together from the same applications(s), and work to at least combine those into a single stored procedure with several SQL statements and some logic. If you implement SQL procedures on V8, the switch to DB2 9 native SQL procedures is a snap - drop and create and go, with no change to the code.
So, go, create, reuse, and be happy. :-)[Read More]
We now have the capability in DB2 9 for z/OS to search text data that is stored in DB2 for z/OS using SQL statements. Wahoo!
You mean you missed the announcement?
And you just followed that link and still couldn't find it? It's under "utilities", no, it's not that kind of utility, but still, that's where it is.
What is added is built-in functions for contains() and score(), and also shipping a text search server which runs on a separate, non-z/OS server. For more details, see the announcements!
One prerequisite for this is to have a WLM application environment set up to run a java user-defined function. The early customers I've been working with have had the most stumbling with this part of it. So, this is something you can set up even if you are not quite to DB2 9 for z/OS yet. I'll post some more on the setup steps for that.
So, what kind of data are you going to search, and what kinds of searches are you going to do?[Read More]
We made a change in DB2 9 for z/OS in order to better package java(TM)code. We now ship DB2 java code such as that required for our XML schema registration and text search password encyrption, to be installed in an HFS/ZFS directory like /usr/lpp/db2/db2910_base
If your installation lets SMP/E default to that directory, then the same set up you use for Java stored procedures in DB2 for z/OS V8 will continue to work. But if you change that, then you need to set a new ENVAR in your JAVAENV dataset such as "DB2_BASE=/usr/lpp/db9a/db2910_base" so we can find our code. Otherwise, you'll see this error when the WLM-SPAS tries to start up: java.lang.NoClassDefFoundError: com.ibm.db2.dsnx9.JARLoader
I know, there are an awful lot of "moving parts" to setting up for running java stored routines. You need the DB2 universal java driver, the z/OS JVM, and JCL and a JAVAENV dataset. The stored procedures redbook has a good chapter on setup. It's a complex environment, but a very powerful one, too![Read More]
I found this article
online today, which highlights the importance of enterprise search.
Company networks contain mountains of structured and unstructured data archived in numerous formats, some of them decades old and stored in secure servers.
IBM also is building a portfolio of enterprise search tools and services, under the OmniFind brand.
Of course you know that DB2 for z/OS data contains mountains of information! This is what our just-released text search support addresses for DB2 for z/OS data - character, binary, and XML. And it's built on OmniFind technology. With this support, you can do text search queries using the built-in CONTAINS() function. It's provided with DB2 9 for z/OS and the no-charge accessories suite.
Now, I know that this is just one piece of enterprise search. In fact, I joke with my colleagues that all of the work that we've put into this is "just an SQL statement". :-) But hey, it's an important piece - it can keep the DB2 for z/OS data where it is and "let the searches come to us".[Read More]
Among the 'recommended practices' that I often present on DB2 for z/OS stored procedures is this one:
- No more than 512 SP's in a WLM
Let me explain why I recommend this. It's actually at the bottom of the list, and that's because it doesn't come up that often. But it has, and when it does, it can cost in I/O. DB2 has a Language Environment table of load modules in each stored procedures address space. For stored procedures defined STAY RESIDENT YES, we only have room for 512 load modules in that table. A load module has to be in the table in order for DB2 to invoke it. So, starting with the 512th, we'll delete it from the table after we call it, even if it's STAY RESIDENT YES. And come to think of it, we have separate tables for TYPE MAIN and TYPE SUB.
So to be completely accurate, the recommendation could actually say something like this:
- No more than 512 different load modules for STAY RESIDENT YES SP's in a WLM application environment, that are all either PROGRAM TYPE MAIN or PROGRAM TYPE SUB and invoked during the lifetime of a single instance of a WLM-SPAS.
For that last bit, remember that different invokers of a stored procedure that end up classified in different WLM enclaves will not have their SPs run in the same instance of a WLM-SPAS.
What's a WLM-SPAS? It's what I use to abbreviate a "WLM-established stored procedures address space".
And this post has motivated me to get a more recent copy of my stored procedures recommended practices presentation out online![Read More]
Also among the 'recommended practices' that I often present on DB2 for z/OS stored procedures is this one:
- Don't call the metadata stored procedures
Many invocations of DB2 for z/OS stored procedures come from a Java(TM) or a CLI application. The software stack for these programs accessing DB2 for z/OS is through a "driver" program. These driver programs have SQL packages bound to DB2 for z/OS, and in the case of the application invoking a stored procedure, there is a fair amount of code executed in the driver program.
For a CLI program (the term CLI is often used interchangeably with ODBC) -- this is usually something running from a Microsoft(TM) application accessing DB2 for z/OS. The DB2 connect software that includes the driver for DB2 for z/OS has some smarts in it so that if the application is coded using incorrect data types for the stored procedure being invoked, the driver recovers and invokes the SQLPROCEDURECOLS metadata stored procedure on DB2 for z/OS to find out what the data types are and then re-sends the stored procedure call to DB2 for z/OS. Yes, you got it right, this means that a poorly coded application can invoke 3 stored procedure calls for every SQL CALL it's trying to do -- one to the original SP, one to SYSIBM.SQLPROCEDURECOLS, and then again to the original SP with the correct parm types! How do you recognize this? Well, you could run a client-side DRDA trace and it will show up there. Or you can look at statistics at the server. Or you can set the value DESCRIBEPARAM=0 in the db2cli.ini file on the client, and let the applications get the error SQLCODE -301 because now the driver won't do the metadata PS call and instead will let the application fail due to using the wrong datatype. Same result if you issue a -STOP PROCEDURE (SYSIBM.SQLPROCEDURECOLS) ACTION(REJECT) command on the DB2 for z/OS server.
For a Java(TM) program, the current driver is the DB2 Universal Java Driver, and it will not invoke the metadata stored procedure. So this is an excellent reason to switch to the current driver, because the older version of the driver went through the CLI code path and had the same problem as described above.
Note that if you invoke a stored procedure from the command line (the CLP), that code will always invoke the SQLPROCDURECOLS stored procedure since the command line doesn't provide anything for what data type the arguments are.
Now, if you are stuck with a CLI program that you can't modify, what can you do to improve the performance of SQLPROCEDURECOLS? Well, APAR PK57017 just shipped which reduces the size of the package for this stored procedure, so you can free up some EDM pool usage and get a small CPU usage improvement. You can also be sure you run RUNSTATS so that the data access for this SP is the most efficient it can be. I have also heard rumors of some customers creating additional indexes on the tables used by SQLPROCEDURECOLS, but I don't have any specifics on that, sorry.[Read More]
When I describe native SQL procedures in DB2 9 for z/OS, I often hear variations of these types of questions:
- Doesn't the external WLM-managed infrastructure provide some throttling of stored procedures? What's going to happen when this is gone?
- Can DBM1 handle the same amount of concurrent stored procedures as multiple WLM-SPAS?
- User routines only use below the bar storage, so how much below the bar storage is available in DBM1 for these native SQL procedures?
In order to answer this, I have to explain a little bit about how DB2 handles native SQL procedures. They are simply packages, with "runtime structures" for the SQL statements to be executed. So, when you invoke a native SQL procedure, DB2 finds and loads the package and executes the statements.
In contrast, an external stored procedure with SQL needs a complete language environment for the user program, and then that external program comes back to DBM1 to get its package loaded and SQL statements executed. That's what needs to be "throttled" - the external program execution environments and their associated TCBs. When an incoming stored procedure request is queued for WLM, the DB2 thread is suspended in DBM1. Many customers have experienced delays and DBM1 storage problems when their WLM goals weren't adjusted properly and the queued requests built up. The solution is to either adjust the WLM goals, or else adjust the limit on DB2 threads (local and/or distributed).
With native SQL procedures, the thread will just switch packages when the call statement is processed and run the procedure - no queuing. The storage used for the local variables is above the bar and managed with efficient algorithms. The maximum concurrent first-level native SQL procedures is effectively the same as your setting for maximum DB2 threads. (What I mean by first-level is that a native SQL procedure may have a nested call to another native SQL procedure, so the actual number of concurrent native SQL procedures may be even higher).
So, I guess the way I'd answer the questions is:
- Yep. When it's gone, SPs will run much more efficiently
- Yep - in fact likely more
- n/a - SQL procedures aren't "really" user routines - they are a pre-defined set of SQL statements, and they don't use below the bar storage
Of course I recommend that you test your native SQL procedures in your environment and measure for yourself, and do capacity planning based on the results of your testing. Native SQL procedures will use some DBM1 storage, after all, and how much depends on what statements and what variables are used in the program.
Oh, and if you didn't recognize it, the "What, me worry?" is a reference to the signature quote from Alfred E. Neuman. It's more than a little tongue-in-cheek.[Read More]
Announced today: New pricing options for DB2 for z/OS running new workloads! All you data center folks who lament to us that pricing for "other" databases can't be compared to DB2 for z/OS - rejoice!!
Announcing today, and already found here is this gem of a news item tidbit:
IBM is also announcing the immediate availability of DB2 for z/OS Value Unit Edition, which provides a new one-time-charge offering that enables the deployment of new application workloads. This offering strengthens the role of System z as a cornerstone for key business initiatives such as SOA, Data Warehousing, Business Intelligence and packaged applications such as SAP. DB2 for z/OS Value Unit Edition and IBM Information Server enable System z clients to further deliver trusted information for their dynamic warehousing requirements.
Just updated: Here is where you can find the gory details.
Is this cool or what? Doesn't this just remove the last and final objection that the application architects have for leaving DB2 for z/OS out of the running for those new applications?
Now, lest you think I am somehow reflecting a non-developer perspective, look, I have spent most of my efforts in DB2 for z/OS developing the kinds of new technologies designed to attract new workloads, and since even I have heard the pricing objection, isn't it perfectly fair for me to mention this in my DW space? And heck, since I am a developer, not a pricing person by any stretch of the imagination, if this has gotten my attention, you know it's big news!
Bring on those new workloads! And then come to us in development and tell us what you need to bring more work onto z, OK?