[plug] dd speed estimates and /dev/hdx grepping

[Derek Fountain]

> If my ragged assumptions are true, I think a block size which is
> half of the size of the cache in the drive you are writing to
> would allow for some swing buffering. Otherwise use the cache
> size. If you have the time, try both of these and see what difference
> it makes.
>
> If my assumptions overlook Linux mucking about with the block
> lengths in transit to the disk then real performance diverges
> uncertainly :-)

I'd have thought the kernel would efficiently buffer the writes and send them 
to the destintation disk in one big hit. Sequential sector writes would be 
easy meat for the elevator, nullifying the effect of a write cache.

AFAICS, the only thing which the kernel and on disk buffering won't interfere 
with is the disk reading, so the best strategy seems to be to use a very big 
block size, slurp in as much data as possible each time, then let the kernel 
figure out how it's going to get it onto the disk.

-- 
> eatapple
core dump

_______________________________________________
plug mailing list
plug at plug.linux.org.au
http://mail.plug.linux.org.au/cgi-bin/mailman/listinfo/plug