Fuzzy Grouping: Any Success With &&> 3 Million Records?
May 18, 2006
I have tried to process > 3 million Fuzzy grouping records on two different servers with no success. 3 mill works but anything above 4 mill doesn't. Some background:
We are trying to de-dup our customer table on: name (.5 min), address1 (.5 min), city (.5 min), state (exact). .8 overall record min score.
Output includes additional fields: customerid, sourceid, address2, country, phonenumber
Without SP1 installed I couldn't even get a few hundred thousand records to process
Two different servers - same problems. Note that SSIS and SQL Server are running locally on both
The higher end server has 4GB RAM, the other 2.5 GB RAM. Plenty of free disk space on both
SQL Server is configured to use 2 GB of RAM max
The page file is currently at 15GB
After running a number of test on both servers trying different batch sizes etc. the one thing I noticed is that it seems to always error out when SSIS takes over and starts chewing up all the available RAM. This happens after the index is created and SSIS starts "warming caches". On both servers SQL Server uses up about 1.6GB of RAM at this point while SSIS keeps taking over RAM until all physical RAM is used up.
Some questions:
Has anyone been able to process more then 3 million records and if so what is your hardware configuration?
Should we try running SSIS from a different server so it has access to the full amount of physical RAM? (so it doesn't have to fight for RAM with SQL Server)
Should we install Win 2003 Enterprise Server so we can add more RAM?
Any ideas why switching to the page file might be causing errors?
I have a requirement to delete 1 Million records from a table having 10 Million data and it's being queried on 24/7 basis (don't have a downtime). how can I achieve that?
I managed to get fuzzy grouping working. The relevant output (_key_in and _key_out) are stored in a new table that is a copy of the old table + fuzzy grouping columns.
How do i get SSIS to store the _key_in and _key_out in the original table? The new matching column _key_out refers to the new key: _key_in. How could i get SSIS translate that to a matching column that refers to my original key?
Hi - we have been evaluating using Fuzzy Grouping and Lookup for maintaining our large list of customer records. Initial testing with Grouping on about 300K records went great but now with a larger sample of 7.3 million records we are running into problems. It doesn't appear to be system limitation - the index is built reasonably quickly and without errors but when it starts the matching we get these errors:
[Fuzzy Grouping Inner Data Flow : DTS.Pipeline] Error: The ProcessInput method on component "Fuzzy Lookup" (86) failed with error code 0x8000FFFF. The identified component returned an error from the ProcessInput method. The error is specific to the component, but the error is fatal and will cause the Data Flow task to stop running.
[Fuzzy Grouping Inner Data Flow : DTS.Pipeline] Error: Thread "WorkThread0" has exited with error code 0x8000FFFF.
[Fuzzy Grouping Inner Data Flow : DTS.Pipeline] Error: Thread "WorkThread1" received a shutdown signal and is terminating. The user requested a shutdown, or an error in another thread is causing the pipeline to shutdown.
[Fuzzy Grouping Inner Data Flow : OLE DB Source [1]] Error: The attempt to add a row to the Data Flow task buffer failed with error code 0xC0047020.
[Fuzzy Grouping Inner Data Flow : DTS.Pipeline] Error: Thread "WorkThread1" has exited with error code 0xC0047039.
[Fuzzy Grouping Inner Data Flow : DTS.Pipeline] Error: The PrimeOutput method on component "OLE DB Source" (1) returned error code 0xC02020C4. The component returned a failure code when the pipeline engine called PrimeOutput(). The meaning of the failure code is defined by the component, but the error is fatal and the pipeline stopped executing.
[Fuzzy Grouping Inner Data Flow : DTS.Pipeline] Error: Thread "SourceThread0" has exited with error code 0xC0047038.
One thing we did find is that our test server didn't have SP1 installed and that seemed to help a lot (we were getting buffer errors prior to SP1). One other note - the desination table is populated with all the data but no scoring has been applied to it.
Does anyone have any ideas what could be causing this?
I have an Oracle table called "Party" which contains Party_Id as primary key and have Party_Name, Party_Addr etc., as fields. We have lot more duplicate party details such as (party_name and party_addr) in this table. We are trying to aviod duplicates using FUZZY logic of SSIS.
1. Is any body suggest me how to create package to avoid duplicates using Fuzzy logic for this scenario(Step by step instructions are good for me to understand SSIS).
2. Could you please provide me some samples for FUZZY(Please send me a sample to my email)
I was running a Fuzzy Grouping task on SQL Server Enterprise Edition SP1 without any issues. I then applied SP2 and now that same Fuzzy Grouping is causing a minidump and terminating the process.
First, does anybody know anything about this kind of issue?
Second, I tried to run the minidump file in Visual Studio but I cannot actually run the dump file in Visual Studio as I keep getting the following exception:
Debugging information for 'DtsDebugHost.exe' cannot be found or does not match. No symbols loaded.
Finally, I did obtain a random error on the server itself that displayed the GUID: 58FC39EB-9DBD-4EA7-B7B4-9404CC6ACFAB.
This GUID appears to be tied to a Dr. Watson error but, again, I cannot figure out what process is breaking.
We do not have any Address Cleansing tools and the requirement is we have to cleanse the data, finding the best possible record which has all info and update other records accordingly.
I am Not sure we can do this Fuzzy Grouping Transformation.
I have been struggling with this for quite awhile so any help would be appreciated.
I need to know if there is away to populate the fuzzy grouping control dynamically. I know you programmatically design a package and customize it in C# but for our purposes we would like to control the SSIS package via database settings. When the settings change the package would then act different. Its a simple a package consisting of an Input - fuzzy grouping - conditional split - output. The connections are setup dynamically using parameters, expressions and a script task. Is there anyway I could do a similar thing for Fuzzy Grouping?
We have a SSIS package which includes Fuzzy Grouping in Data Flow. It takes two columns from source table and saves outputs in different table with match score etc. Following is the way we are doing it: 1. Load required data from table using OLEDB connection (source) 2. Sort the data 3. Apply Fuzzy grouping (using dedicated database instead tempdb and MinSimilarity = 0.6) 4. Send to destination table using OLEDB connection (destination)
In input table we have millions of records. It takes too long to execute and even sometime it fails after running 12 hours. Any suggestions for performance improvement are welcomed.
I have a few questions about the amounts of resources used by the fuzzy grouping transformation. I am running a little less than 5mil records through a fuzzy grouping that exact matches one column and fuzzy matches one. The server executing the package is a dual-core xeon with 2gb ram, running a default instance of sql 2005 enterprise.
I have been attempting to execute this package for a while now but it keeps erroring out for various reasons. At first, it was from a lack of available memory. I limited the memory usage of sql server to 256mb and set the buffer temp storage path, which alleviated those errors. However, now, my tempdb transaction log is growing significantly. It failed once for not being able to grow and reallocate quickly enough, but enlarging the auto-growth factor fixed that. Then, it filled up the volume the tempdb log was on, so now I have moved it to the san and am about to try again.
I was wondering, does anyone have a general idea on approximate resource usage by fuzzy grouping? Specifically, is there an approximate relation between the number of records grouped and the amount of ram/pagefile required? Also, on the database backend, how big can I expect the tempdb data/log files to get?
I need some advice on fuzzy lookup / grouping design. I have a requirement that, I think, is between lookup and grouping transformations.
In one of our applications, users can enter manually a label for some information in the database. Every month, I will store all the new data in our OLAP DB, and I want to group these labels with a fuzzy logic. Historical data (already loaded) have to be grouped, as well as new data coming every month.
I have no predefined canonical data, so Fuzzy Lookup seems not adapted to my pb. Fuzzy Grouping seems ok, but it would require to put historical data as well as new data as an input of the Fuzzy Grouping Transfo to constitute groups. This seems not efficient to me.
My question is how to calculate the similarity by using SQL query, example LIKE % , order by.....? Now i'm doing a function same like fuzzy grouping but i do not know how to get the answer, mean how they get match with those selected row of data.
Hope my question is clear. How to write the correct query? What should i do? I 'm newbie in Integration Services, so i need ur explaination in step by step if there hv correction.
I am looking forward to hearing from you shortly and thanks a lot in advance.
I have recently decided to dedupe my data but i am having a problem after running fuzzy grouping with the query on updating which duplicate to keep
_key_in is unique, _key_out is the duplicates so for example:
_key_in , _key_out , name , score , dedupe 1 , 1 , ron , 10 , purge 2 , 1 , ronn , 15 , keep 3 , 3 , john , 5 , keep 4 , 4 , matt , 15 , keep 5 , 4 , mat , 10 , purge 6 , 4 , matt , 15 , purge
I want to keep the _key_out with the higher score by setting the field de_dupe to 'keep' and the remainder to 'purge'. The score can also be the same within a duplicate so in the case it is the same i just need to keep one it doesnt matter which one. The query i have below nearly works but it marks duplicates with the same score as keep.
Code: UPDATE b SET b.dedupe_result = 'keep' FROM [BusinessListings].[dbo].[MongoOrganisationACTM1Destination] b INNER JOIN
I've seen one other post on this topic from October 2005 and I thought I'd bring it up again. I've a Fuzzy Grouping component in my data flow. The output data from it appears to be the result of records spliced into other records. This includes pass-through columns, not merely "clean" or similarity columns. For example (I've added the suffixes for illustrative purposes):
I was wondering how Fuzzy Grouping deals with and handles first name similarities. Is there a way to configure it so that Anthony = Tony, Bill = William, etc€¦? I created a simple package with several rows containing similar first names and ran the fuzzy grouping on the first name column. I received only one possible duplicate of Will = William which was at 56%. I lowered the threshold down to 1% and still only one match.
Now I understand and appreciate the reasons for this but was wondering if this type of situation was considered and a way of dealing with it is available.
Is there a way the fuzzy lookup or grouping can be trained so that similarities and confidence values rely on previously matched strong links?
For example: I can link 80% of my two datasets using one strong identifier (say phone #) which I trust. My goal then, is to use the probability of matching of the rest of my linking fields (say Name,Address,Gender,DOB) in a "matched by phone number" pair to train a fuzzy lookup task to be done on the unlinked 20% of the datasets.
This "training set" would in theory influence the similarity and confidence values of the fuzzy output since each linking column would carry a different weight or contribution towards a confident match.
Does anyone out there knows how to do this in practice in SSIS?
Will the fuzzy grouping task match a null value to an empty string (or spaces)? I've got 5 columns I'm matching on, and one of them may be null for certain rows but an empty string for others. Given the 4 other columns may match, will this difference stop similar columns being grouped together?
(Someone's modified my grouped data since it was deduped, which takes a while, and I'm hoping for a quick answer on this).
I have a table that I need to identify similarities so I'm running a Fuzzy Grouping Process. I'm getting the follow errors and I can't identify the problema since all the fields are varchar, except for the first that is int but not use in the fuzzy.
select MSSEndCustomerTPID , orgname , address1 , cityname , statename , countryname from [sales].[vw_Fact_VolumeSales] a inner join [GMOFBI].[dbo].[vw_Dim_MSS_Organization] b on a.EndCustomerOrganizationKey=b.MSSOrganizationKey
I am currently working on a simple page to insert 1.6 million UK postcode records into an SQL server table. The table has three columns for the postcode, longditude coordinate and lattitude coordinate. The data is sourced from a pipe (|) delimited txt file and inserted into the database using a FOR loop. The problem I have is that the page will hang after inserting only 10,000 records, the page displays either an invalid View State error or a page cannot be found error. Now I assume the viewstate error stems from the fact that there is a form on the page which simply contains a button to execute the script and a few labels to show the progress. But without the form and associated viewstate the insert still fails to complete.... any ideas?? Would I be better running this on a thread or should I just do it in stages and be patient. I have now modified the page to read the database on load and pick up from where it crashes?
I have a table that has 4+ million records. I need to update those records. I am facing some performance issue. Can someone please advice?
update stage set batch_status = 1 where update_status = 0
Update transaction Set aId = s.aId, b = s.b,
from stage s Where s.aId = transaction.aId and s.batch_status = 1
Update stage Set update_status = 1, batch_status = 2
where
batch_status = 1
When I run the above query with "set rowcount 1000", it runs in one minute. When I run the query for "set rowcount 10000", it runs in 1 hour 56 minutes. Can someone help me to optimize it?
Hey folks...So I have a table that looks like this:CREATE TABLE [tblStation] ([CAMPAIGN] [varchar] (8),[LISTNUM] [varchar] (10),[PHONE] [varchar] (10),[EVENTTIME] [datetime] ,[STATION] [int],[OPERATOR] [varchar] (16),[EVENTCODE] [varchar],[CALLSPAN] [decimal](18, 0),[FDISP] [int],[RECORDNUM] [varchar],[STC] [varchar],[PROMOC] [varchar],[EXP_CAMP] [varchar],[PROMO3] [varchar],[MAXATT] [char],[LISTNAME] [varchar],[SITENAME] [char],[Row_id] [int] IDENTITYIt's taking nine seconds to run the following command:SELECT count([fdisp])FROM [TrunkFiles_new].[dbo].[tblStation] WITH (NOLOCK)WHERE fdisp IS NULLAnyone familiar with a table of this size having performance likethis? The [fdisp] column has a non clustered index on it.Thanks in advance...
How well SQL Server can support 300 million records... Any body is working on big database like this. can anyone give me some input on this. it's going to be 60GB database size.
i have a directory database with approx. 80 million records. i am feeding the database with bulk_insert. Indexing one of the fields took about 8 hrs. After indexing when i run queries with the indexed field the response time is under 1 sec. However if i run select queries with like on non-indexed fields it takes more than 2 mins. So i decided to index 4 other fields in the database and it looks like the indexing process is going to run for 2 days. i am a novice in SQL database design and i am not sure if this is the best way to index the table. i am just using create index. Any suggestions / advice welcome.
Hello, What is the fastest way to update 20million records in our database. I have tried to do a simple update statement like this: update trail_log with (tablockx, holdlock) set trail_log .entry_by = users.user_identity from users where trail_log.entry_by = users.user_id
but it take 10 plus hours to run since it cannot commit the transactions until the very end. So was was thinking that I need to commit in batch like after 50K but that is slow as well. Set rowcount 50000 Declare @rc int Set @rc=50000 While @rc=50000 Begin Begin Transaction update trail_log With (tablockx, holdlock) set trail_log.entry_by = users.user_identity from users where trail_log.entry_by = users.user_id and trail_log.entry_by not like '%[0-9]%' Select @rc=@@rowcount --Commit the transaction Commit End go I have let the above statement run for 1.5 hours and it only update 450000 rows. Any ideas... Maybe I'm doing it wrong. Please Help!!
I have a sql script that updates records in a table with 40 million records.
There is some functionality in the script that could be put away in functions for code reuse/elegance.
Functions would cause execution overhead.
What else could I use besides functions that would allow me the code reuse and not compromise the execution over head? Is there any thing like includes in TSQL that would allow me to do so?
I have a new client with an existing system that has just over 2 million business listings in one table. Each business listing is associated with one business category.
* Company Table (around 20 fields):
companyID companyName categoryID state postCode etc.
* Category Table (5 fields)
categoryID categoryName etc.
We are using MSSQL 2005 Express Edition with Advanced Services
A free text search needs to be performed on the companyName and categoryName limited by region (state and or postcode).
1) What kind of response times should I expect for the free text search (I have not used the free text search before)
2) How should I index the companyName and categoryName so they are both used in a joined query? i.e. Do I just configure the free text search index on each field separately and it should work?
I want to compare ONLY 1 Column values from 2 tables having more than 4.9 million records. There is a difference of 4000 rows between the 2 tables.
SELECT ID From TABLE1 where ID not in (SELECT DISTINCT ID From TABLE2)
My above query took nearly 4.5 hours to run and I had to cancel it. Is there a better way to write the query . I just want to compare the ID - column values which are missing in TABLE2
I come from a web based world were loading 1.5 million records into a temp table is suicide. I’m doing more data warehouse stuff now and I was looking into optimizing a buddies proc and noticed he was loading 1.5 million records into a temp table. We had a discussion about it because being from a web world I was drastically against it. He on the other hand didn’t feel it was an issue being it gets called once maybe twice a day. The tempdb is set to autogrow and it is on a different drive than all the other databases on the box. It has one ldf and mdf. He’s creating an index on the table after load. Why we shouldn’t be loading 1.5 million recs into temp table?
