Re: Long-running queries

gukal@cc.gatech.edu (Sreenivas Gukal) wrote: >Now for the thesis issues :-) A couple of recent research papers >try to solve this by using transient versioning. The basic >idea here is that every update creates a new version. The queries >access the prior versions while the transactions update the >current versions. Storage overhead is reduced by making the prior >versions transient; they are removed when the queries accessing >them complete. My work goes on similar lines, but also shows >that the storage overhead for transient versions can be further >reduced. I think this sounds like something we all use every day, the Oracle Rdb snapshot files. This work fine as long as the long transactions are not very long, and not very many. If you have a lot of very long read transactions going on at the same time as extensive update processing is done by a lot of users, you can often end up in a situation where your snapshot files (or transient versions if you prefere to call them so) is several times as large as your files containing the data. This is my experience from practical work. If you use a replicate database, you only need two times as much space as the size of the data (I count indexes, metadata and other kind of overhead as "data" here). Name: Magnus Weiman Company: Datalogikonsult AB http://DECUS.SE/~weimanm mailto:WEIMANM@DECUS.SE S-mail: Box 52, S-590 79 VRETA KLOSTER, SWEDEN Phone: +46 13 131330 Fax: +46 13 61142 GSM: +46 70 7440806 minicall text: 393662