[texhax] [tex-live] problems with tl2016 pretest

[Zdenek Wagner]

2016-05-14 17:47 GMT+02:00 Jim Diamond <Jim.Diamond at acadiau.ca>:

> On Sat, May 14, 2016 at 09:32 (+0200), Zdenek Wagner wrote:
>
> > 2016-05-14 0:14 GMT+02:00 Karl Berry <karl at freefriends.org>:
>
> >> it just may be an issue of the design of tl-install, in that it
> >> doesn't overlap "download the next file" with "unpack this file".
>
> >> Doing the operations (install package A while downloading the next
> >> package B) concurrently is indeed beyond the scope of the
> >> implementation as it stands.
> Understood.
>
> >> And it doesn't sound easy, or reliable, to accomplish :(. -k
> > It is very difficult.
> In principle, I find that surprising.  But if you mean "without a
> significant re-write of the current implementation, it would be
> difficult to do it", I'm happy to believe you.
>
> > You have packages of different size
> I don't see the relevance of that (except if someone designed a
> program which would only work if all the packages were the same size,
> which seems like a very curious design).
> > and users have different bandwidth.
> Nor that.
> > In theory, you may quickly download several small packages if you
> > have a gigabit line and it may cause concurrent unpacking and
> > installation of several packages, so you may end up with many
> > processes which will mean that yout tcp/ip subsystem might discard
> > incoming packages due to overloading of the processor.
> I think your concerns are specific to one class of designs.  But as
> you already realized, as evidenced by your next sentence,
> > It could be solved by having one process for downloading that would
> > put the names of downloaded packages to a queue and another process
> > will unpack and install them.
> other (superior) designs exist.
>
> > If you have Intel i5 or better, it will work but if you install to
> > Raspberry Pi, it may cause problems due to limited resources.
> Again, I think your concerns may be founded upon one particular
> design, as opposed to whether fundamental problems exist.
>

Mere download from the nearest mirror takes less than 1 minute, unpacking
is considerably slower although in my computer the whole /usr/local lives
on a SSD (the full installation takes about 20 minutes). I have not tried
but without clever IPC I will easily end up with more than 10 concurrent
processes which is two much for quad core.

>
> > Moreover, I am not sure whether the same IPC is available for all
> > platforms, so such a solution is really difficult and will need a
> > lot of testing and debugging.
> I have never tried to install texlive on anything but Linux systems,
> so I don't know how much commonality there is between
> install-tl-unx.tar.gz and the windows version.  If a lot, then that is
> a concern, assuming that maintaining the commonality is seen as a good
> thing.  Mind you, I would assume that the existence of the given Perl
> libraries on all machines of interest would take care of OS-specific
> concerns.  (But I say that as someone who avoids programming in Perl.)
>

I have never tried portable IPC.

>
> Cheers.
>                                 Jim
>



Zdeněk Wagner
http://ttsm.icpf.cas.cz/team/wagner.shtml
http://icebearsoft.euweb.cz
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://tug.org/pipermail/texhax/attachments/20160514/80c2a163/attachment.html>