location of 'datasets'

[PeteM2]

Does having the dataset descriptor file is on a different file sytem than the data files cause any problems?

The reason i am asking this is that the descriptor and data files are currently on the same file system and in order to reduce I/O queuing i would like to keep the descriptor file for new datasets on the current file system and move the data files onto a number of different file systems.

[qt_ky]

It should not cause any problems as long as all the file systems involved are available when needed. The *.apt config file(s) are flexible in allowing you to point to various paths across nodes.

[jwiles]

No, that scenario is quite common, especially in shared cluster/grid environments.

However, do not just simply move existing data segment files to new storage. You will need to either copy the existing datasets using a datastage job or the orchadmin command or simply recreate the datasets.

Regards,

[PeteM2]

Does the config file node 'dataset' file system only relate to where new data files should be located?

Therefore is it the case that the location of existing data files would be derived from the decriptor file without any reference to the config file?

[jwiles]

Dataset descriptor files contain the location of all data segment files which contain the data for that dataset. This is why you cannot simply move the data segments to another location. No, you cannot modify the dataset descriptor file...it is a proprietary binary format.

Yes, the disk resource entries in the configuration file determine where data segments files will be written per logical node, as documented in the Parallel Job Developer's Guide and mentioned many times in the forum.

Regards,

[qt_ky]

Data set descriptor files (or control files) contain a copy of the config file that was used at the time of data set creation.