Your input file having 10000 records, if the job went down at 9000 record you need to restart from 1st record onwards in this case system resources are waste. For avoiding this scenario checkpoint concept will come. They will define check points for some interval (for each 500 records like) for above job fail no need to go for 1st record for restarting go for last checkpoint(i.e 8500) and restart job for last checkpoint.
