Join IIUG
 for   
 

Informix News
18 Nov 13 - ZDNet - Top 20 mobile skills in demand... Read
09 Sep 13 - telecompaper - Shaspa and Tatung have shown a new smart home platform at Ifa in Berlin. Powered by the IBM Informix software... Read
06 Sep 13 - IBM data magazine - Mission Accomplished - Miami, Florida will be the backdrop for the 2014 IIUG Informix Conference... Read
01 Feb 13 - IBM Data Magazine - Are your database backups safe? Lester Knutsen (IBM Champion) writes about database back up safety using "archecker"... Read
14 Nov 12 - IBM - IBM's Big Data For Smart Grid Goes Live In Texas... Read
3 Oct 12 - The Financial - IBM and TransWorks Collaborate to Help Louisiana-Pacific Corporation Achieve Supply Chain Efficiency... Read
28 Aug 12 - techCLOUD9 - Splunk kicks up a SaaS Storm... Read
10 Aug 12 - businessCLOUD9 - Is this the other half of Cloud monitoring?... Read
3 Aug 12 - IBM data management - Supercharging the data warehouse while keeping costs down IBM Informix Warehouse Accelerator (IWA) delivers superior performance for in-memory analytics processing... Read
2 Aug 12 - channelbiz - Oninit Group launches Pay Per Pulse cloud-based service... Read
28 May 12 - Bloor - David Norfolk on the recent Informix benchmark "pretty impressive results"... Read
23 May 12 - DBTA - Informix Genero: A Way to Modernize Informix 4GL Applications... Read
9 Apr 12 - Mastering Data Management - Upping the Informix Ante: Advanced Data Tools... Read
22 Mar 12 - developerWorks - Optimizing Informix database access... Read
14 Mar 12 - BernieSpang.com - International Informix User Group set to meet in San Diego... Read
1 Mar 12 - IBM Data Management - IIUG Heads West for 2012 - Get ready for sun and sand in San Diego... Read
1 Mar 12 - IBM Data Management - Running Informix on Solid-State Drives.Speed Up Database Access... Read
26 Feb 12 - BernieSpan.com - Better results, lower cost for a broad set of new IBM clients and partners... Read
24 Feb 12 - developerWorks - Informix Warehouse Accelerator: Continuous Acceleration during Data Refresh... Read
6 Feb 12 - PRLOG - Informix port delivers unlimited database scalability for popular SaaS application ... Read
2 Feb 12 - developerWorks - Loading data with the IBM Informix TimeSeries Plug-in for Data Studio... Read
1 Feb 12 - developerWorks - 100 Tech Tips, #47: Log-in to Fix Central... Read
13 Jan 12 - MC Press online - Informix Dynamic Server Entices New Users with Free Production Edition ... Read
11 Jan 12 - Computerworld - Ecologic Analytics and Landis+Gyr -- Suitors Decide to Tie the Knot... Read
9 Jan 12 - planetIDS.com - DNS impact on Informix / Impacto do DNS no Informix... Read
8 Sep 11 - TMCnet.com - IBM Offers Database Solution to Enable Smart Meter Data Capture... Read
1 Aug 11 - IBM Data Management Magazine - IIUG user view: Happy 10th anniversary to IBM and Informix... Read
8 Jul 11 - Database Trends and Applications - Managing Time Series Data with Informix... Read
31 May 11 - Smart Grid - The meter data management pitfall utilities are overlooking... Read
27 May 11 - IBM Data Management Magazine - IIUG user view: Big data, big time ( Series data, warehouse acceleration, and 4GLs )... Read
16 May 11 - Business Wire - HiT Software Announces DBMoto for Enterprise Integration, Adds Informix. Log-based Change Data Capture... Read
21 Mar 11 - Yahoo! Finance - IBM and Cable&Wireless Worldwide Announce UK Smart Energy Cloud... Read
14 Mar 11 - MarketWatch - Fuzzy Logix and IBM Unveil In-Database Analytics for IBM Informix... Read
11 Mar 11 - InvestorPlace - It's Time to Give IBM Props: How many tech stocks are up 53% since the dot-com boom?... Read
9 Mar 11 - DBTA - Database Administration and the Goal of Diminishing Downtime... Read
2 Feb 11 - DBTAs - Informix 11.7 Flexible Grid Provides a Different Way of Looking at Database Servers... Read
27 Jan 11 - exactsolutions - Exact to Add Informix Support to Database Replay, SQL Monitoring Solutions... Read
25 Jan 11 - PR Newswire - Bank of China in the UK Works With IBM to Become a Smarter, Greener Bank... Read
12 Oct 10 - Database Trends and Applications - Informix 11.7: The Beginning of the Next Decade of IBM Informix... Read
20 Sep 10 - planetIDS.com - ITG analyst paper: Cost/Benefit case for IBM Informix as compared to Microsoft SQL Server... Read
20 Jul 10 - IBM Announcements - IBM Informix Choice Edition V11.50 helps deploy low-cost scalable and reliable solutions for Apple Macintosh and Microsoft Windows... Read
20 Jul 10 - IBM Announcements - Software withdrawal: Elite Support for Informix Ultimate-C Edition... Read
24 May 10 - eWeek Europe - IBM Supplies Database Tech For EU Smart Grid... Read
23 May 10 - SiliconIndia - IBM's smart metering system allows wise use of energy... Read
21 May 10 - CNET - IBM to help people monitor energy use... Read
20 May 10 - ebiz - IBM Teams With Hildebrand To Bring Smart Metering To Homes Across Britain... Read
19 May 10 - The New Blog Times - Misurare il consumo energetico: DEHEMS è pronto... Read
19 May 10 - ZDNet - IBM software in your home? Pact enables five-city smart meter pilot in Europe... Read
17 March 10 - ZDNet (blog) David Morgenstern - TCO: New research finds Macs in the enterprise easier, cheaper to manage than... Read
17 March 2010 - Virtualization Review - ...key components of Big Blue's platform to the commercial cloud such as its WebSphere suite of application ser vers and its DB2 and Informix databases... Read
10 February 2010 - The Wall Street Journal - International Business Machines is expanding an initiative to win over students and professors on its products. How do they lure the college crowd?... Read

End of Support Dates
IDS 11.10 - 30 Sep 2012
Subscribe to notifications

IIUG on Facebook IIUG on Twitter

[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

IDS Forum

Re: Question about sysindexes.clust

Posted By: Art Kagel
Date: Tuesday, 18 November 2008, at 4:28 p.m.

In Response To: Re: Question about sysindexes.clust (Obnoxio The Clown)

OK Clown, you asked for my analysis, you got it. There are:

69million rows
rowsize ranges from 155 to 244 bytes (e to a varchar column) (8-12 rows fit
per page on 2K pages) assume an average of 10 rows per page
Estimated number of data pages: 7 million
Estimated number of index pages for table qty_det for the selected index:
410,000 with FILLFACTOR = 100

FILLFACTOR = 50: 820,000 pages of index (168 keys per page: 2020/
(4byte key + 4byte rowid + 50% internal overhead))

Those facts in mind we can analyse what may be happening during this query
before and after CLUSTERING or sorting the incoming data which accomplishes
the same thing as long as the data load is single threaded.

Searching the index for this table for the estimated 1,094,000 rows of table
qty_head which match the selection criteria (I'm assuming that the
optcombo_qty table adds little filter value since it is searched AFTER the
join between qty_head and qty_det) would require reading from a minimum of
6,500 index pages for table qty_det if ALL 1.1million matching rows had
sequential serial numbers to 820,000 pages if the serial values are evenly
distributed across the 69 million value range. Then if the rows are not
stored on the same data pages there may be as many as 1.1million data pages
to be read (minimum ~100,000 pages). That would require a cache of at least
106,500 pages just to hold the data and index keys for this one table plus
additional space for the other two tables in the query and for other queries
- and a maximum of 2.2 million pages. Since originally the data was likely
to have been fairly randomely distributed across the disks due to prior
deletions and VARCHAR row relocations, the number of pages needed in cache
will tend towards the upper limit of 2.2 million rather than the lower
limit. I assume that the cache on this server is not in excess of 2.5
million pages.

Now, if the rows are sorted into serial number order when loaded (or after
CLUSTERING), then the data rows that we are most likely to be interested in
(assuming we are interested mostly in the most recent rows added to the
table) are all contiguous on disk as are their keys in the index. That
means the number of disk pages needed to be read into cache (and the chances
that a page needed for more than one row has to be read more than once due
to timing out of the cache during the search) are minimized to only 106,500
pages as compared to the worst case - that the rows we are interested in are
scattered all over the disk due to the data not being sorted.

Add to this the possibility that since these rows contain a VARCHAR column
that may have grown over time, there may be many of these rows that required
two IOs to fetch into memory not one before any reorganization. Now the
actual timings show that the actual distribution of data on disk was far
from worst case, but also far worse that the best case.

So, back to the original question: Do we watch the
sysindices/sysindexes.cluster column to determine when a table should be
reorganized? Honestly and embarrasedly, no. Most of us have fallen away
from that particular optimization. I used to look at this in 4.0 and 5.0
days, but have not concerned with it for many years. One reason is that it
is a problem that tends to affect a small subset of queries (indeed in your
case it was killing only one particular SELECT statement) so it is one of
the last things I check for if I cannot improve a stubborn query any other
way. Also, other reorg opportunities tend to minimize the impact of this
problem as we try to minimize the number of extents in very large tables.

Art

On Tue, Nov 18, 2008 at 3:01 PM, Obnoxio The Clown
<obnoxio@serendipita.com>wrote:

> Eric Rowell wrote:
> > The following is the explain plain for the only query in the program
> which
> > isn't running correctly. The first is the plan from the normal run
> (filter
> > data had to be changed to project my job). The info after that is from a
> > diff of the explain outs for 2 more configurations. Also during all tests
> > we used a production like system (smaller CPU and slower SAN) but we know
> > the approx. modifier from prod to dev. When we sorted the data by the
> > serial_link and reloaded (changing the fragmentation as shown earlier
> things
> > run much better.
> >
> > EXPLAIN from normal run...
> >
> > QUERY:
> > ------
> > SELECT qd.option_list, oq.optid
> > FROM qty_head qh, qty_det qd, optcombo_qty oq
> > WHERE ((qh.set_no = "FINDA" AND qh.version = "09") OR
> >
> > (qh.set_no = "ORFND" AND qh.version = "01"))
> >
> > AND qh.serial_key = qd.serial_link
> >
> > AND qd.serial_key = oq.option_group_id
> >
> > AND qh.dept_code IN ("AAA","BBB","CCC","DDD","EEE")
> >
> > AND qh.community = "00"
> >
> > Estimated Cost: 1049986
> > Estimated # of Rows Returned: 1083718
> > 1) informix.qh: INDEX PATH
> >
> > (1) Index Keys: dept_code community set_no version phase_no lot
> > area_group_id
> >
> > (Key-First) (Serial, fragments: ALL)
> >
> > Lower Index Filter: (informix.qh.dept_code = 'AAA' AND
> > informix.qh.community = '00' )
> >
> > Index Key Filters: (( (informix.qh.set_no = 'FINDA' AND
> > informix.qh.version = '09')
> >
> > OR (informix.qh.set_no = 'ORFND' AND
> > informix.qh.version = '01')))
> >
> > (2) Index Keys: dept_code community set_no version phase_no lot
> > area_group_id
> >
> > (Key-First) (Serial, fragments: ALL)
> >
> > Lower Index Filter: (informix.qh.dept_code = 'BBB' AND
> > informix.qh.community = '00' )
> >
> > Index Key Filters: (( (informix.qh.set_no = 'FINDA' AND
> > informix.qh.version = '09')
> >
> > OR (informix.qh.set_no = 'ORFND' AND
> > informix.qh.version = '01')))
> >
> > (3) Index Keys: dept_code community set_no version phase_no lot
> > area_group_id
> >
> > (Key-First) (Serial, fragments: ALL)
> >
> > Lower Index Filter: (informix.qh.dept_code = 'CCC' AND
> > informix.qh.community = '00' )
> >
> > Index Key Filters: (( (informix.qh.set_no = 'FINDA' AND
> > informix.qh.version = '09')
> >
> > OR (informix.qh.set_no = 'ORFND' AND
> > informix.qh.version = '01')))
> >
> > (4) Index Keys: dept_code community set_no version phase_no lot
> > area_group_id
> >
> > (Key-First) (Serial, fragments: ALL)
> >
> > Lower Index Filter: (informix.qh.dept_code = 'DDD' AND
> > informix.qh.community = '00' )
> >
> > Index Key Filters: (( (informix.qh.set_no = 'FINDA' AND
> > informix.qh.version = '09')
> >
> > OR (informix.qh.set_no = 'ORFND' AND
> > informix.qh.version = '01')))
> >
> > (5) Index Keys: dept_code community set_no version phase_no lot
> > area_group_id
> >
> > (Key-First) (Serial, fragments: ALL)
> >
> > Lower Index Filter: (informix.qh.dept_code = 'EEE' AND
> > informix.qh.community = '00' )
> >
> > Index Key Filters: (( (informix.qh.set_no = 'FINDA' AND
> > informix.qh.version = '09' )
> >
> > OR (informix.qh.set_no = 'ORFND' AND
> > informix.qh.version = '01')))
> > 2) informix.qd: INDEX PATH
> >
> > (1) Index Keys: serial_link (Serial, fragments: ALL)
> >
> > Lower Index Filter: informix.qh.serial_key = informix.qd.serial_link
> > NESTED LOOP JOIN
> > 3) informix.optcombo_head: INDEX PATH
> >
> > (1) Index Keys: serial_key option_list option_count (Serial,
> > fragments: ALL)
> >
> > Lower Index Filter: informix.qd.optcombo_head_key =
> > informix.optcombo_head.serial_key
> > NESTED LOOP JOIN
> > 4) product.qty_detail: INDEX PATH
> >
> > (1) Index Keys: serial_key (Serial, fragments: ALL)
> >
> > Lower Index Filter: informix.qd.serial_key =
> > product.qty_detail.serial_key
> > NESTED LOOP JOIN
> > 5) informix.oq: INDEX PATH
> >
> > (1) Index Keys: optcombo_head_key opt_id (Key-Only) (Serial,
> > fragments: ALL)
> >
> > Lower Index Filter: informix.oq.optcombo_head_key =
> > product.qty_detail.optcombo_head_key
> > NESTED LOOP JOIN
> >
> > Diff of Explain Plan after just reloading the table:
> > < Estimated Cost: 1049986
> > < Estimated # of Rows Returned: 1083718
> > ---
> >> Estimated Cost: 2348476
> >> Estimated # of Rows Returned: 1486409
> >
> > Diff of Explain Plan after reloading the sorting data (using
> serial_link):
> > < Estimated Cost: 1049986
> > < Estimated # of Rows Returned: 1083718
> > ---
> >> Estimated Cost: 851019
> >> Estimated # of Rows Returned: 1094319
> >
> > The cost difference appears to be in line with the change to the
> > sysindex.clust value for the indexes.
>
> And which table do you query from to see if there are missing rows and
> insert them?
>
> Basically, I can't disagree with your analysis, I'm just curious as to
> why ordering should have such a measurable impact on performance given
> the explain plan. In general, it doesn't.
>
> I feel like there is something lurking in the nether hells of my brain,
> but I can't quite drag it out. Last time I saw something like this, the
> root cause was "XXXX in my opinion" but I can't remember what "XXXX" was.
>
> Give me time. Or Art will be along shortly. :o)
>
> --
> Cheers,
> Obnoxio The Clown
>
> http://obotheclown.blogspot.com
>
>
>
> *******************************************************************************
> Forum Note: Use "Reply" to post a response in the discussion forum.
>
>

--
Art S. Kagel
Oninit (www.oninit.com)
IIUG Board of Directors (art@iiug.org)

Disclaimer: Please keep in mind that my own opinions are my own opinions and
do not reflect on my employer, Oninit, the IIUG, nor any other organization
with which I am associated either explicitly or implicitly. Neither do
those opinions reflect those of other individuals affiliated with any entity
with which I am affiliated nor those of the entities themselves.

Messages In This Thread

[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

IDS Forum is maintained by Administrator with WebBBS 5.12.