M
martinz
Guest
Hi all, We're cleaning up a database for one of our customers. It's a production database, and off course downtime should be as short as possible. As it's a large database for us (around 1.5 TB), and some of the tables can be completely removed, and we want to reclaim unused diskspace, we decided to dump and load the database. We can do that per area. After dumping and loading, we need to do an idxbuild. This takes 8 hours for an area with about 163 GB of data. This area only contains tables, indexes are in a separate area. The command used is proutil _db_ -C idxbuild area _indexarea_ -TB 64 -TM 32 -T _tmpdir_ -thread 1 -datascanthreads 2 -mergethreads 2 -B 5000 -i The machine used is a linux machine with 4 cpu's and 32GB of memory. We have one volume for the database, and some other volumes for the tmpdir. All are on a SAN with SSD storage (I know, local storage would be much better, but we can't get that right now). The database is a OpenEdge 11.5.1 database. Scan for the index area takes about 4 seconds, as it is empty. Scan for the data area takes about 2.3 hours. After that, most of the time seems to be going into sorting index groups. Any pointers for speeding up the idxbuild? Spreading out the sortfiles over multiple volumes using -SS and a .srt-file? Using -TF to use more memory? Using more extents for the index- or data-area? Regards, Martin.
Continue reading...
Continue reading...