Re: [PATCH] sort: parallel external sort implementation

[Chen Guo]

Hi Ralf,

> > > +      --threads=N           use no more than N threads to improve

> parallelism\n\
> > 
> > Threads is a bit of an implementation detail?
> > Perhaps it would be better to use --parallel so one
> > could move to multiprocess in future if desired?
> 
> --jobs is already used by make, so why not reuse that, if you decide to go 
> with
> a command-line argument?  That's pretty neutral in terms of how the 
> parallelism
> is implemented internally, too.


I dislike semantic arguments but I'm afraid I have to make one here... Using
make to compile a few files results in make using gcc or whatever to compile
each file individually. In sort however, assuming the input files fit in 
buffer, they'd
all be read in as one large continuous buffer and operated on together, thus
making "jobs" a bit of a misnomer.