[R-gui] Re: [R] Feedback about SciViews?

[Prof Brian Ripley]

On Thu, 1 May 2003, Luke Whitaker wrote:

> On Wed, 30 Apr 2003, A.J. Rossini wrote:
> 
> > "Liaw, Andy" <andy_liaw at merck.com> writes:
> > 
> > >> The GUI itself (i.e., how it operated, what menus were where, etc.) 
> > >> was fine, but it was completely useless for anyone sitting at a 
> > >> remote host, due to dreadful image quality and poor performance when 
> > >> displaying anywhere other than on the console of the machine on which 
> > >> SPlus was actually running.
> > >
> > > AFAIK the problem is the X server: it gives you a black window if the X
> > > display is 16-bit.  If you can live with 8-bit display, the Java GUI will
> > > work.  It's that way on all Unix (X?) platform.
> > 
> > I thought Java on Linux would do 8 and 24, but not 16?  (or something
> > truly weird like that).
> > 
> > But that doesn't address performance issues, which still can be pretty
> > bad.
> 
> It seems the problem here is limited to running R remotely. 

No (and were we not discussing the *S-PLUS* Java GUI?). S-PLUS Java will
not start locally on Solaris in 96Mb of RAM (and my home Sun only has
64Mb).  In reality you need much more, as you need memory to do the
calculations too.

I've just fired it up on my new Dual Athlon 2600 with 1Gb RAM under RH8:
I'd say the GUI was about the speed of S-PLUS for Windows on a 133MHz 
Pentium, that is noticeably slow.  I have seen faster Java GUIs elsewhere, 
but none has approached the speed of a native GUI, and until recently they 
have been between irksome and intolerable.

> Given
> the power of cheap consumer desktop machines these days, this seems
> to me less of an issue than it used to be. Expensive commercial
> software probably gets run remotely more than it needs to because
> of licensing issues, which obviously don't apply to R.

In my experience R is often run remotely to share machines with large 
amounts of memory.  We don't routinely put 1Gb in people's desktops.

> If the GUI can be cleanly split from the compute engine of R, then

I think that is lot harder than you suppose, as the S model is inherently 
of sequential calculations on objects.  People expect GUIs to multitask.

> it should be quite easy to get the GUI running locally even if the
> compute engine is remote - although I re-iterate my question for the
> need to run free software remotely. My very modest desktop machine
> is far more powerfull than the mainframe that I learnt to use SAS
> on, and not so far short of the very expensive file server that I
> use now.

That's the perspective of someone in a first-world company (I guess from
the email address).  R is designed to run under 16Mb RAM on say 133MHz
machines, because some people using it really do have machines like that.
Our students often have machines with 128Mb or less.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595