Login ProductsSalesSupportDownloadsAbout |
Home » Technical Support » DBISAM Technical Support » Support Forums » DBISAM General » View Thread |
Messages 1 to 6 of 6 total |
Slow update when using transactions |
Mon, Aug 7 2006 12:26 PM | Permanent Link |
Michael Binette | I have a sql script that does an update on a large table using an un-
optimized where clause. If I don't use transactions then it completes in 7 seconds. If I use a transaction it takes 13 minutes! Is there anything I can do? -- Thanks, Michael Binette |
Mon, Aug 7 2006 12:29 PM | Permanent Link |
Michael Binette | I just tried changing the index so that it WAS optimized and it was still
slow. |
Mon, Aug 7 2006 12:47 PM | Permanent Link |
Arnd Baranowski | Michael,
Yes, don't use a transaction. Queries implicitly use optimized restricted transactions on the updated table (have a look at chapter 2.12 SQL and Transactions of the manual). I don't think that you will get any better performance if you place transactions on your own. Arnd Michael Binette wrote: > I have a sql script that does an update on a large table using an un- > optimized where clause. If I don't use transactions then it completes in 7 > seconds. If I use a transaction it takes 13 minutes! > > Is there anything I can do? > > -- > Thanks, > Michael Binette |
Mon, Aug 7 2006 2:52 PM | Permanent Link |
Tim Young [Elevate Software] Elevate Software, Inc. timyoung@elevatesoft.com | Michael,
<< I have a sql script that does an update on a large table using an un-optimized where clause. If I don't use transactions then it completes in 7 seconds. If I use a transaction it takes 13 minutes! Is there anything I can do? >> Arnd is correct. UPDATEs already use transactions for the commit batches during their execution, and their done in an optimal way. If the UPDATE is interrupted due to an improper shutdown, etc. then just run the UPDATE again and it will begin where it left off before. -- Tim Young Elevate Software www.elevatesoft.com |
Tue, Aug 8 2006 9:06 AM | Permanent Link |
Michael Binette | "Tim Young [Elevate Software]" <timyoung@elevatesoft.com> wrote in
news:70C7C0CA-A259-4574-98AF-2CAE0A0CD9AD@news.elevatesoft.com: > Michael, > > << I have a sql script that does an update on a large table using an > un-optimized where clause. If I don't use transactions then it > completes in 7 seconds. If I use a transaction it takes 13 minutes! > > Is there anything I can do? >> > > Arnd is correct. UPDATEs already use transactions for the commit > batches during their execution, and their done in an optimal way. If > the UPDATE is interrupted due to an improper shutdown, etc. then just > run the UPDATE again and it will begin where it left off before. > Yes but my update script contains 3 update statements and I also have a table that is being posted to. That is why I am using a transaction. I want it all to either pass or fail. It doesn't do me much good if it fails in the middle. Then I have to try to determine at which step it failed at. |
Tue, Aug 8 2006 9:39 AM | Permanent Link |
Tim Young [Elevate Software] Elevate Software, Inc. timyoung@elevatesoft.com | Michael,
<< Yes but my update script contains 3 update statements and I also have a table that is being posted to. That is why I am using a transaction. I want it all to either pass or fail. It doesn't do me much good if it fails in the middle. Then I have to try to determine at which step it failed at. >> In that case you'll have to deal with the performance impact of the transaction caching a very large number of updates at one time. -- Tim Young Elevate Software www.elevatesoft.com |
This web page was last updated on Friday, April 19, 2024 at 07:09 AM | Privacy PolicySite Map © 2024 Elevate Software, Inc. All Rights Reserved Questions or comments ? E-mail us at info@elevatesoft.com |