This is a summary of responses to my question regarding Splus for linux
vs solaris on intel hardware. Based on this response it looks like we
will give linux a shot. Many thanks to all who responded!
Original posting:
Roger> We have always run Splus on Suns here, but have just received
Roger> a gift from Intel of a dell poweredge 6300 server, dual
Roger> processor, xeon chip, etc,etc. So we need to make a decision
Roger> about what to do with this machine. My original inclination
Roger> was to run linux, S+ 5.0, and R. But I would like to know
Roger> whether there is an option to run solaris. I didn't see
Roger> anything about intel/solaris/Splus on the Splus webpages.
Here is the "semi-official" response from Chuck Taylor at statsci:
There is no version of Splus that runs under Solaris on an Intel platform,
nor is there one in the works. Similarly, there is no version of Splus
that runs under Linux on any platform other than Intel (SPARC or DEC
Alpha). Here is a summary of platform requirements for Splus 5.0:
Sun Solaris 2.5 or newer
Linux (Intel) kernel 2.0.31 or newer
We are working on future releases of Splus 5.0 for DEC Alpha OSF1, HP-UX,
SGI IRIX, and IBM AIX.
We used the RedHat version of Linux on Intel for in-house development, and
have also tested on SuSE Linux. RedHat 5.2 would be a good choice.
This is from Brian Ripley:
I think S-PLUS on linux is a bit rough at the moment, a bit rougher
than on Solaris, but time will likely change that.
This is from Doug Bates:
As far as I know there are no versions of S-PLUS for Solaris/Intel.
It would be of great interest to me if there were because we run a lab
of such machines. I have tried on occasion to convince Mathsoft to
port S-PLUS to Solaris/Intel but have not been successful. From
experience with other software, we expect that it would not be
necessary to make any changes in the code base - it would just be a
matter of recompiling on same code used for SPARC to get the
Solaris/Intel version.
We are caught in a bit of a squeeze with those machines. We are able
to run Minitab (version 9) on them. Minitab, Inc. was willing to try
recompiling their Solaris/SPARC version for Solaris/Intel. It worked
without a hitch. We are also able to use the AFS file system on those
machines. However, we cannot use S-PLUS nor can we use many other
common programs distributed as binaries. Adobe acroread is one such
program. I think Matlab was another. Both of these have Linux
versions. You have to realize that, to many software companies,
Solaris means Solaris/SPARC.
If we ran Linux on those computers we would have access to other
commercial applications such as S-PLUS and to a wider range of
open-source software. We would probably have to give up Minitab but
that would not be such a great loss. Minitab, Inc. is only developing
the Windows version as far as I can see. They are are version 12 on
that platform but have not updated their Unix offerings past version
9. More importantly for us, AFS support would not be as good as it is
on Solaris.
Considering the current momentum behind Linux and the S-PLUS support,
I would choose Linux over Solaris/Intel. We are not switching
ourselves because our systems people don't want to add yet another
operating system to the supported systems list.
I think Solaris/Intel is a peculiar product for Sun. They want it to
be successful but not too successful. That is, they want to have
people adopting Solaris as the operating systems but they would prefer
to sell them SPARC computers and the operating system rather than just
selling them the operating system. I could foresee a time when they
would drag their feet on incorporating changes into Solaris/Intel that
they make in Solaris/SPARC.
R runs fine on both Solaris/Intel and Linux.
This is from Todd Taylor
Solaris is not an option. For reasons I guess I'd need a
marketing degree to understand, Statsci has never ported its
Sun/Solaris Splus to Intel/Solaris. Linux *is* an option, or at
least will be when Splus is finally released for it.
We have also always used Splus on Suns. I recently participated
in the beta testing of Splus 5.*, specifically so I could test
and compare the Intel/Linux platform with Sun/Solaris. I found
that on hardware that was roughly equivalent in raw
number-crunching power and memory, Splus/Intel/Linux
outperformed Splus/Sun/Solaris. Performance was similar until
the problem became large (i.e., caused substantial swapping).
To my surprise, the Intel/Linux box was sometimes twice as fast
under those conditions. I'm not sure why---maybe because Linux
is smaller, maybe it has better swapping algorithms, maybe its
compiler had better libraries, who knows. And I don't know if
the results would scale up for *really* large problems or for
lots of simultaneous users.
I also personally prefer Linux to Solaris, at least for a
workstation, for various reasons (it's smaller, faster,
generally comes with more software, compiles new open-source
software with less hassle, etc.).
url: http://www.econ.uiuc.edu Roger Koenker
email roger@ysidro.econ.uiuc.edu Department of Economics
vox: 217-333-4558 University of Illinois
fax: 217-244-6678 Champaign, IL 61820
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news
|