Just to chip in, I've had a few complaints about this as one of the guys behind malariacontrol.net.
The basic formula for the volume of data being saved to disk per hour is s*f*n where s is the size of a checkpoint, f is the frequency (num written per hour), and n is the number of jobs active in parallel. Obviously* *keeping *s* small is the job of the project (we try, but there is a limit),* n* is up to the user, but *f* is specified by the client/user and AFAICT the client default cannot be specified by the project (we had one guy complain that several computers using 1000MBit/s NAS were totally choking the network). Short version of the story is that increasing the checkpoint interval (even quite drastically since checkpoints are only useful should the application crash in a non-deterministic fashion or be killed unexpectedly) should help, but I don't know what else can be done. -Diggory
