Swapoff, anyone tried?

version: v21.03.2
Compressed data size before ingestion: 375GB
Total out size: 894GB
Zero: 32 core, 256 GB, 2TB SSD
Run: 31 hours for map phase, 11 map reduce phase

Bulk loader command

dgraph bulk -f /coldstartinput/upload/pending_predicates -s /coldstartinput/upload/rdf_schema/patient.rdf --out /coldstartoutput/out --replace_out --num_go_routines=20 --reducers=7 --format=rdf --store_xids --map_shards=20 --reduce_shards=10 > check.log &
  • Bulk slows down dramatically towards the last 30%
  • I can see too much swapping happening

Question:

  • Does swapoff help here?
  • How to turn off swapoff ? Note sure if the cloud operators allow this?