[SATLUG] simultaneously burning multiple DVDs

Craig sargonemail at gmail.com
Sun Jan 4 15:27:39 CST 2015

Looks like the 3.0-3.8 kernels are not the way to go for high i/o :

On Wed, Dec 31, 2014 at 2:02 AM, Craig <sargonemail at gmail.com> wrote:

> *Borries Demeler* on  *Tue Dec 30 11:35:29 CST 2014 *wrote:
> >  If I needed 2 MB on board memory buffer, does that ninux cannot supply
> > the data fast enough to the drive if I use more than one drive?
> >
> > If so, the question comes up why Linux cannot do that. Assuming I use a
> RAM drive
> > as the source for the iso to be written to multiple drives, is it
> because the
> > RAM drive is too slow to be read?? That I find hard to believe.
> > So is it because the data cannot be moved fast enough through the system
> bus?
> > With Gbit bus speeds, I also find this hard to believe.
> > Is it because the CPU is too slow? I have multiple cores, each could run
> > a separate process, so this can't possibly be the reason either.
> >
> > I am still stumped why this doesn't work as expected.
> Are you absolutely sure the ram drive is non-swappable?
> If the ram drive or portion there of gets sent to disk swap (e.g. the
> portion not yet burned,
> but not in use), then needs to be swapped back in from disk (or portions
> there of) when it's time
> to be used, actual physical ram read access might not happen in time for
> the buffer to be sent data
> in time to  prevent buffer underrun.
> -- particularly if swap is on a platter drive and/or other things are
> jockying for use of the remaining
> hard ram (e.g. swapped between swap space & physical ram).  Obviously in
> this case, DVD data transfer
> will be slower/lower priority than swapping between physical ram / virtual
> ram.
> Even if the ram drive is marked as "do not swap out of physical ram",
> without a real time kernel,
> swapping programs between physical ram & swap space can easily create
> enough timing issues
> which would destroy any advantages gained by the non-swappable ram drive.
> So, having a ram drive may actually make things worse.
> A non-real time kernel does not guarantee availability of resources at a
> fixed point in time.
> ** try opening multiple tabs under a web browser, where each tab has a
> youtube or flash media link.
>     How many does it take before the active tab video starts skipping?
> (take a look at vmstat / available memory & swap usage)
>     How many tabs does it take before it becomes a chore to switch between
> tabs? (take a look at iostat e.g. swap usage! ).
>     How many open/active tabs before the web browser crashes under the
> strain?

More information about the SATLUG mailing list