AnsweredAssumed Answered

Splitting a PI system into two

Question asked by tinklerj on Mar 13, 2019
Latest reply on Jan 13, 2020 by Roger Palmen

I have done a fair bit of merging PI systems, but now I am faced with splitting one system (53,000 tags) into two systems of approx 20,000 and 5,000 tags respectively (non-overlapping).  The 28,000 surplus tags are being retired.  The key point is that the archive files on each new system must not contain any data associated with tags on the other system (even if inaccessible because the tag does not exist in the point database on that system).  Therefore I have to reprocess all archives (1500, 1TB total in mostly 1GB archives) for each new system, to produce archives containing no data from the other system.


Clearly I need to clone the original PI DA to two new systems, and on each delete unwanted tags from the point database.


For data purging I can see two possibilities:

1. Copy backups of all source archives to each of the target cloned/pruned PI systems.  Reprocess all archives, on each, without special options, and register.  I assume that data records for missing tags will simply not be copied to the output archive (silently), rather than generating errors?

2. Create ID conversion file for each new PI system containing only the IDs of tags for that system.  Reprocess archives (or backups of archives to avoid unregistering originals) twice, using each ID conversion file in turn.  This could be done on the source (unsplit) PI DA system and the output files copied to the cloned/pruned systems for registering.



Option 1 has the advantage of performing reprocessing for each target in parallel so only taking half the time.  Does anyone have benchmarks for how long reprocessing a 1GB archive with 50,000 tags takes?

Option 2 requires less disk space as the source archives do not need to be copied anywhere.  I assume the two sets of output archives combined will take up half the space of the single set of source archives, since only half the tags are being preserved.


For option 2, would all this reprocessing seriously impact the performance of the source PI DA system which is a production server?