Posted Feb 4, 2010 2:40 UTC (Thu) by NCunningham (guest, #6457)
Parent article: Improving readahead
I have similar issues in TuxOnIce. The approach I've used, which might also be useful here, is (among other things) to throttle readahead based on the rate at which data has been read so far. Perhaps a throughput-based limit might be worth a look here, too? (Particularly in the context of Jens' objection).