[Rd] Performance issues with R2.15.1 built from source on Solaris

[Prof Brian Ripley]

On 10/08/2012 17:50, Eberle, Anthony wrote:
> Yes, the T4 physical host that the zone is configured on is SPARC.  Are
> you saying that this is a function of the hardware or would possibly
> updating to a newer version of gcc help?  I'm actually going to see if I
> can build a newer version of gcc anyway just to eliminate that variable.

It is a function of the hardware.

You will likely do better (I do on a T2) with the native compiler 
(Solaris Studio) than any version of gcc on Sparc, but things involving 
long doubles are going to be slow.  However, they are a very small part 
of a typical mix of R work (if e.g. what CRAN packages do is typical).

In any case, I would suggest using a recent version of gcc, as R-patched 
does not compile with gcc3/g77: we have finally required a more modern 
Fortran (and there are C things like inlining that require more C99 
compliance than gcc3 has).


> I picked the "sd" and "R-Benchmark" tests as I am new to R and
> functioning more as an administrator than a modeler or statistical
> engineer.  I too have seen that some operations do perform better on
> certain systems than others.  This makes sense as I know different
> hardware and CPU's are better at some things than others.  However I'm
> not familiar enough with R yet to know what the functions are doing
> under the hood and which would be more performing on what hardware.
>
> However I will try to check into the specs for the T4 and see if
> anything is mentioned about the "long double" or something along these
> lines.  Thanks.
>
> Anthony
>
> -----Original Message-----
> From: r-devel-bounces at r-project.org
> [mailto:r-devel-bounces at r-project.org] On Behalf Of Radford Neal
> Sent: Friday, August 10, 2012 9:49 AM
> To: r-devel at r-project.org
> Subject: [Rd] Performance issues with R2.15.1 built from source on
> Solaris
>
> Is this a SPARC system?  On at least some SPARC systems, the "long
> double"
> type in C is implemented very slowly in software, and it seems that it
> is used for the sums done when calculating standard deviations with
> "sd".
>
>     Radford Neal
>
>
>> Date: Wed, 8 Aug 2012 18:55:37 -0500
>> From: "Eberle, Anthony" <aeber at allstate.com>
>> To: <r-devel at r-project.org>
>> Subject: [Rd] Performance issues with R2.15.1 built from source on
>> 	Solaris?
>
>> I have a question about building R (2.15.1) from source on a Solaris
>> 10 64bit virtual server with 2 cores and 16GB memory that is running
>> on an Oracle T4 server.  Based on several tests I have done this
>> configuration has been several orders of magnitude slower than any
>> other configuration I've tested.
>>
>> A simple test of some code to calculate the standard deviation 10000
>> times (simple code to consume resources) takes on average 121.498
>> seconds on the Solaris server where the next worst system (Redhat
>> Linux) takes 1.567 seconds:
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>


-- 
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