Batch.Indexing

BATCH.INDEXING is a routine which permits the buffering of index transactions in memory until certain pre-set limits are reached. Using this routine, speed improvements of 30%+ are typical. The usage is simple, just call

     BATCH.INDEXING(1, File_Var)

to turn batching on, and

     BATCH.INDEXING(0, File_Var)

to turn it off.

From thereon, index transactions are buffered in memory for that file and not actually written to the ! datafile. The pre-set limits are documented in the latest "System Subroutines" book from Revelation Technologies as follows

"BATCH.INDEXING either buffers a set amount of data (1,000 bytes) or buffers for a limited time (5 seconds), whichever occurs first, so you need not be concerned with overrunning memory space".

The information buffered by BATCH.INDEXING is only flushed to disk by the ! code and by a direct call to BATCH.INDEXING turning the buffering off. Further, only one file is buffered at a time. This has four major implications ;

• You should NEVER logoff without first turning BATCH.INDEXING off. If you do, any buffered transactions will be lost. (Fixed in 2.12)
• The time between flushes is only 5 seconds if records are continually being written to disk through the ! code. That is, if a record is written to disk and then the machine is left standing for 24 hours the buffered transaction will still be in memory. However the moment a record is written to the same file the record will be flushed.
• Neither INDEX.FLUSH nor background indexing flush the buffered transactions to disk, regardless of the "Delay before indexing" settings.
• If BATCH.INDEXING is started on one file and transactions are buffered and then BATCH.INDEXING is started on a new file, the old transactions will be written to the new file.

(Volume 3, Issue 9, Pages 5,6)