1 sec, elapsed or cpu ? if elapsed, measure can be wrong due to processor load > 95%.
Assume you can run in confortable situation to avoid mis-measures : check with WRKSYSSTS
- processor < 80 %
- not too much database page faults. no value to suggest, depends too much of context
- 0 act-inel
what is now the record run delay ? changed ?
1 sec is very too long. How many UNIQUE index implied in update ? all necessary ? if you update more than 70% of records think to rebuild index after running the prog.
in some case, OVRDBF NBRRCDS(…) can improve (this improvement may conflicts with unique index)
Are you sure you read each record only one time ? it’s a crazy way, but it’ a way of analyze
About CHAIN, SETLL+READ=CHAIN. if there is a difference, it is not remarkable.