Minutes of the GSE UK IMS Working Group Meeting - Manchester University, 1^st July 1999

Chairman: Frank Fleming ( Frank.Fleming@barclays.co.uk)

Secretary: Peter Armstrong ( peter_armstrong@exchange.bmc.com) - 01276-419835

Administration

• Introduction
  • Welcome to one and all, including the DBCTL people. Bit of a sparse turnout - don't know why. Numbers vary a lot at our meetings. Best excuses - attending a wedding, AD2K. Worst excuse was from Alan and Dougie, who are running an e-seminar in Europe for two days, which clashes with this meeting.

  • Frank handed out a little welcome pack as he couldn't make an agenda foil (problems with copier), so copies distributed. Pack contained agenda; attendees and apologies: description and objectives of the GSE UK IMS Working Group - can also be found at web.ukonline.co.uk/dougie.lawson/gse; Steering Committee members; details of other GSE UK Working Groups; calendar of meetings (ours and other groups) Frank is aware of, and a feedback form. This feedback form is not designed to be used for assessing the quality of the presentations - what we are interested in is the relevance of the presentations and ideas for future meetings - please think about this and fill in ideas. We are always struggling for user presentations. Feedback form must be handed in.

  • Thanks to Ian for organising agenda and having his holiday nearby so he could attend.

  • Frank ran through a quick outline of the agenda.

  • The registration form was passed round with a request to update details, especially email ids.

  • In order to attend these meetings you have to be a member of GUIDE SHARE Europe - a couple of attendees need to sort this out.

• Minutes
  • Everybody had received. Minutes and agenda are out on Web (to be found at http://web.ukonline.co.uk/dougie.lawson/gse).

• Apologies
  • List attached.

• Correspondence
  • GSE Journal issue 1 (dated June 1999). A flyer for the IBM UNIX Solutions Technical Conference - might be useful as you can run UNIX on sensible boxes now. Flyer for 4^th GSE Technical Symposium in Porto, Portugal on October 11-12. Details of the next Large Systems Working Group meeting . Minutes from Storage Management meeting on 10^th July.

• General
  • There was a recent GSE board meeting - twice a year the Working Group chairmen meet together, chaired by Phil Howard-Knight - the new region manager. Major items discussed were:
    • Everybody agreed that it would be a good ides if the Working Groups had a Y2K session in the autumn.
    • GSE UK has 140 members (was 172 last year). Europe has 1500 members. Phil sent out 165 surveys in the UK - 29 returned. Phil will do a follow-up in the 4^th quarter.
    • Requirements - UK system has bad press of late and fallen into disrepute. The GSE end of the system is currently not working well. Last year GSE UK raised 7 requirements, whereas Germany raised ca. 90 (always knew they were a bunch of whingers!!!). There is now an Internet-based system, which will hopefully ease things. Frank will work with Neil on this.
    • How to get presenters - the feedback from has been used and has been helpful. People appear to like to write on a form rather than volunteer in public.
    • There will be no UK region conference this year due to Y2K.
    • Marketing GSE in UK - Frank asked how many people (their companies / their bosses) see these meetings as "jollies" - nobody present. Peter - probably the people who are not here.
    • Next year - Frank is currently talking to the CICS Working Group chairman. They have 2-day meetings with good attendance. One idea is to run a joint IMS/CICS/DB2/MQ meeting.
    • Netview Working Group is now called the Network Systems Management Working Group.
    • Best attendance reported as being IBM Warwick.

• Final meeting of year for this group will be on 8^th October at IBM South Bank.

IBM What's New - Pete Sadler (IBM IMS WorldWide Technical Support)

• UK GSE IMS web site now has link from IMS home page.

• Two documents distributed - the IMS Quarterly and a flyer for the IMS Technical Conference in San Diego. Please take copies. Go to the Technical Conference. Frank - costs less than going to Paris! Next year's European Technical Conference is 27^th to 31^st March in Barcelona.

• No dates set for next year's education classes yet.

• DB Analyser is being stabilised and will not support new IMS features- look at DB Tools - Frank - there are also other products available.

• www.redbooks.ibm.com can be used for online access to red books and also to red pieces, which are early drafts. Be a bit cautious with the red pieces as they haven't been checked / edited yet, but they are still useful.

• There will be a redbook later in the year on the subject of Pete's presentation below.

• A copy of the foils was distributed.

Database Dynamic Allocation - Frank Fleming (Barclays Bank)

• Frank took us through a history of how dynamic allocation came about in IMS. Main problem was that the library had to be APF-authorised which drove people crazy.

• Frank asked where BMC gets its information from for dynamic allocation. Peter - RECON. (Bit added by me later - because we dynamically allocate not only DBs but also logs, Image Copies, Change Accumulates etc.)

• APAR PQ12171 now allows MDA blocks to be placed in a non-APF library.

• Barclays (and other sites) have RESLIB split into various bits, some or all of which are concatenated together in different regions. Have MDA blocks in a separate library. For data sharing they are trying to clone as much as possible, but e.g. MSC can't be cloned. So for issues like this and for maintenance reasons, they have different RESLIBs and NUCLEUSs and program product libraries, but have shared MDALIBs. They don't share any data with batch regions, so no changes required in these regions.

• So what about PQ12171? It allows CTL, DLISAS and batch DLI/DBB regions to access MDA blocks in a non APF-authorised library (IMSDALOC DD statement for CTL and DLISAS, STEPLIB for batch DL/I). So, the simple solution here is to get APF removed for these. The BMC utilities are fine as they work from RECON, but what about DBRC SAS?

• The DBRC SAS has no separate DD statement, so tried leaving MDALIB on STEPLIB. This unfortunately gives a U0623 abend - the problem is that the DBRC SAS STEPLIB still has to be APF authorised, whereas the DBRC utility doesn't.

• So where do I put my RECON MDA blocks? Pete - have you APARed this? Have discussed with Dougie, who thinks it is working as designed, so should we be raising a requirement. Pete will talk to the developer.

• If you put them in the NUCLEUS or RESLIB, then you have duplicate blocks pointing at the same RECONs, with all the associated change control issues.

• Therefore, Frank's suggestion is to use the RECONs for dynamic allocation of all DBs - full function and Fast Path. There was general discussion and the audience agreed as long as it was optional, because of e.g. test systems.

• As far as to where to put the RECON blocks / why have to be APF authorised. Why have different RESLIBs - for maintenance. Santa Teresa says you can start/restart BMPs on different maintenance levels but not on different releases.

• So Frank is going to raise two requirements, one for using RECON and one for not needing APF authorisation for the DBRC SAS, which he will pass to Neil and we will get out a voting sheet with the minutes.

• Is PQ12171 for IMS 6.1 only? Think so, but may be retrofitted?

• A copy of the handouts was distributed.

Business Intelligence from DL/I data - Pete Sadler (IBM UK)

• This is one of the sessions that Pete is giving in San Diego.

• IMS is great for what it is designed for, but it was not designed for business intelligence type applications. It was designed for OLTP, not for OLAP (online analytical processing).

• Pete described the problem, and how you would go about getting at your IMS data / how you would extract it to the format you require it in for mining / analytical processing etc. with all the advantages and disadvantages. Do I need history, do I need trend analysis? What are the performance implications on the mainframe and on the network? Which products are available and how do they work?

• Do I have to archive the log first in DPROP from IMS to DB2? Yes, but how up-to-date must it be? Timer-initiated log switches? This would typically be one hour latency, if you want more recent than that, then you could use e.g. Data Joiner access. By the way, at archive time, the manuals show you how to create a subset of the OLDS (with the records you need) called a CDCDS on disk to avoid mounting SLDS cartridges for the DPROP Selector.

• Pete then went on to describe Datamarts and how to set these up.

• Logging considerations with DPROPR - is it logging the complete row? Yes, will obviously depend on how many data items it is sensitive to.

• You will probably find that a degree of data cleansing is required.

• Minimum IMS levels etc.? Relatively old, no problems.

• Datajoiner relational gateway - impact in IMS? Runs a BMP style program, also working on an OTMA implementation. Beware if your SQL call turns into a "scan the database".

• A copy of the foils was distributed.

Problems forum, hints and tips, contacts for lunch

• Requirements - none.

• Problems
  • Name:Frank Fleming - Barclays
    Problem:We recently implemented LE as our default runtime (taking our old SYS1.COBLIB/COBOLVS or whatever off the LINKLST). We were good boys and tried to implement the use of LRR by putting CEELRRIN as a message region pre-initialisation routine in DFSINTxx as per the IMS V5 Performance Red Book.

    We already had our own pre-init exit (to prevent us running multiple MPR STCs with the same name - it upsets our automation) already installed so we followed the docs (Install vol 2) and put CEELRRIN on the second line of the DXFSINTxx member. But we hadn't spotted info APAR II09411 (point 11) indicating that the docs are wrong and we should have coded KDB345S0,CEELRRIN. Consequently performance went down the tubes.

    This update has not yet hit the manuals.

    Contacts/Comments:Pete - think there is an article off the COBOL home page about implementing LE environment - have a look.

  • Name:Malcolm Shooter - Boots
    Problem:CICS trans and regions - how do you find out which connects to which?

    Contacts/Comments:Neil Price, TNT - DIS A will normally give you tran code. Have to buy 3^rd party monitoring package. Ian - this is still a grey area.

  • Name:Malcolm Shooter - Boots
    Problem:Anyone got OMEGAMON for DBCTL? Is it any good?

    Contacts/Comments: Neil Price,TNT - better than having nothing. Used to be only monitor. Not sure if it gives the link-up between CICS and DBCTL. Not sure about TMON or Mainview. OMEGAMON DBTCL appears to be OMEGAMON IMS with no TM and treats DBCTL as BMPs. Ian - same problems with DC Monitor.

    Ian - can this group send a message to Candle that we are not happy. Boots - DB2 people didn't like Mainview, so went for OMEGAMON. Frank - IMS people didn't like OMEGAMON, so went the other way!!

  • Name:Neil Price, TNT
    Problem:If you have max PST set at 999 and you haven't got UQ21096 (V5), then log archive falls over because it can only handle 998 PSTs.

    Contacts/Comments:

A Black Hole in DBRC - Peter Armstrong (BMC)

• Yet again I have been playing around with DBRC because customers have been reporting strange happenings to me.

• The presentation started by reviewing exactly how DBRC works with batch jobs and batch backout with the records stored in RECON and the actions you are meant to take. I then ran through a series of tests, where I ran some batch jobs and intentionally made them fall over (using a very small output log). I then tried running batch backout from an unclosed log - the good news is that DBRC did not like this, the bad news is that the message it gives you is extremely unhelpful - DATA SET NOT FOUND IN RECON.

• I then ran another series of tests where I not only made the batch job fail but also purposely killed the batch backout job (again using a tiny output log). During this I discovered that DBRC changes the LOG STARTIME in the SUBSYS record of the failed batch job to point at the PRILOG record from the batch backout.

• The problem is that a customer ran this scenario, backout fell over, and then they went and looked in RECON to see if the job had done any updates and if not clean it up - in other words they issued CHANGE.SUBSYS ABNORMAL. Well, the problem is that DBRC now thinks the batch job has done no updates because the SUBSYS record is pointing at the backout PRILOG which has no ALLOCs or LOGALLs yet, so the command works, cleans up RECON and hey presto the DBs are a load of old rubbish. (If this makes no sense, talk to me!!!)

• I then asked the attendees if this made sense and had I found a problem - consensus was yes and Pete Sadler has subsequently talked to IBM about it:

• • • IBM responses:

    • DFS3294A ... DATA SET NOT FOUND IN RECON --
      • OK, the data set *is* in the RECON and the message text is misleading. We can update the message explanation in the next release of the books.

        (Fair enough)

    • LSN reported in the DFS616I message is not recorded in the RECON --
      • Right, the logger does know about LSNs, but we're abending and that's not the time to be making more calls to DBRC. It would be nice if it were stuck in the RECON so that log recovery could automatically get it, but we don't consider this a problem. (Up until V5 RSR, we didn't even put the LSNs in the RECON.) And the LSN in the message is not necessarily the last LSN on the log.

        (As discussed at the meeting - put in X'FFFFFFF' when you close the log is what I would recommend)

    • SUBSYS record incorrect after BBO failure (leading to corrupted DB's) -
      • Now this is a problem. The customer should open a PMR. We will recreate so that we can investigate further.

        (Here's the embarrassing bit - I've forgotten which customer it was, but I am tracking them down and will get this raised)

• The final part of this story is of course a reminder that BMC wrote Batch Control Facility to handle all stupid problems like this in batch, and it works fine and cleans everything up properly and automatically.

• A copy of the foils was distributed.

General Open Discussion

• Question - DBR command issued from CA DADS+ comes back with RC, but nothing found in log. After invetsigation, one theory was that this was caused by a long thread running rather than a short thread - command not executed with long thread is one of the explanations in Messages and Codes. What is a short thread and what is a long thread and can I choose in a DBCTL environment? Frank - BMP is long, MPP short. Ian - all my CICS threads are treated as short even if they run all day. Don't know how to change it. Pete - would expect all CICS threads to be treated as short. Conversational? Don't think that has an effect. Searched on DIAL?

• In days gone by used to search log data sets for various commands - searched logs here for 02 records, but couldn't find any. Perhaps not being logged. Only records recoverable ones. Is there a switch to turn off logging of commands - no. Nothing on MVS log to show it ever got to DBCTL. How does this product know command has failed as it is an asynchronous command? Have you tried Enterprise Assist? Not yet. Have had spurious problems with DADS+ not issuing commands that it says it has. Long discussion on writing home-grown program, how does DADS work etc.

• Discussion on DB-Groups in V6. Do they do FEOV. What if one fails. All commands are asynchronous so you have to check which has worked and which not.

DL/I Options Review - Discussion Topic

How long since you looked at DFSVSAMP? What uses OSAM - all the IMS system datasets. On DIAL - look at IMS Buffer tuning and you will get lots of entries.

• CICS should be top Dispatching Priority, then DBCTL, then DLISAS, then DBRC.
• Any documentation on tuning LSR pool? Use G320-6004 (new one in Pete's list). Use SEAID - VSAM.large buffer pool analysis . Run GTF trace and then into tool, but now out of date.
• Lots of questions on OSAM SB. The SBs are a secondary pool - used for read, not on write. OSAM sequential buffering can speed up series of insert calls as well as GN. This was followed by short discussion of advantages to be gained by turning on SB, and how it started or not depending on sequential level of processing. Pete - this is the sort of stuff that gets presented at the Technical Conferences or SHARE in the US. Next SHARE is in Chicago in August and covers IMS, CICS, DB2 etc. Most SAHRE presentations are available on their website.
• Hiperspace - can we copy DB2 by using this VSAM feature to load small, or medium-sized DBs into memory and remove disk I/O completely? If you want to use hiperspace buffering with batch jobs, you should define the batch region as non-swappable, otherwise the hiperspace buffers will be swapped out with the job, and you will lose the hiperspace benefits.

  Hiperspace buffers must all be >=4K. How many using it? VSAM only. Can also use individual pools and allocate specific DBs to these. We used to use different CI and blocksizes - not needed anymore. Can use subpools, and if they are big, then back them up with a hyperspace. Frank - we use a private pool for one DB, which is small but has a huge hit rate - pool size is the whole DB (40 cyls). Not such a good design when you go to sharing because of buffer invalidation. Now looking at OSAM caching, VSO DEDB or whatever.

• Hardware compression with IMS DBs? Yes, and it can improve performance. Advantage of software compression is that segments are compressed on DASD, in the Contorl Unit, in the buffer pool - only when the program actually needs to look at them do they get expanded, so you can get more in your buffer pool etc. Also makes utilities run faster you are copying / recovering / reorganising less data. Two parts of compression - an engine (hardware microcode is the most efficient) and a control mechanism - usually a software product.

  You can insert SDEPs under h/w compression but not read them back!! SDEP scan does not invoke compression exit. Any DBCTL using DEDB? No. If you can't compress a segment, h/w compression can add 10 bytes and write back How do you turn it on? COMPRTN in DBD. Easy to implement - just control what is compressed so it gets expanded.

• OSAM - multi-volume still an issue.
• GSAM - SMS is allocating automatically with 4 volumes. Have a look at BMC's ARC or SPICE.
• Don't use IDRC on batch logs, because it makes batch backout perform like a dog. Default blocksize in batch is 1K, which is too small. Also don't use IDRC on SLDS in case you have to run backout. Frank - we just give SLDS backout a big region.
• If you are using clever DASD with h/w compression, should I still use s/w compression - yes because you use buffers more effectively. Also doesn't help on 4G limit.
• Compression can increase logging in Fast Path

DFSVSAMP is described in volume 2 of Installation Guide and in Admin Guide DB. Also look in IMS Performance redbook from Alan Cooper. Follow Dallas home page to presentations - you will find Bob Gendry's presentation on tuning buffer pools

AOB

Fill in feedback form - especially ideas and should we meet in Warwick? Can we close down DBCTL subgroup or do we need more info? We will continue to have DB in the main working group if we do this, or have a DBCTL meeting if e.g. there is a new version. Less than 50% of DBCTL people attend the CICS meeting. If you have ideas afterwards - send an email.

Next meetings: 28^th October for IMS group.

GSE UK IMS WG FEEDBACK FORM - RESULTS

Topics

• How to realistically performance/volume test changes to production systems.

• Shared Q Implementation (including EMH?)

• Please talk through IMS gen procedures.

• We don't register any DBs with DBRC. DBCTL is shut down at night whilst batch runs. Is Peter horrified? (YES YES YES!!!) How many sites don't? Why should we if restoring after a crash is acceptable? (Restoring how to what point? - call me and discuss if you want - Peter)

• Online data sharing

• Partitioned DBs

• Internals from professional IBM or OEMs

• Compression

• Peformance Issues related to HDAM, HISAM

• Internet Access & IMS