Can anyone give me a free account on fast SPARC hardware ?
Dave
ask - there is no harm in asking.
Is there anyone that can give me a *free* shell account on *fast* SPARC hardware
to support a major open-source project
I've already got a Blade 2000 with 2 x 1.2 GHz, so it needs to be quicker than
that to be any use.
Sun very kindly donated a 16-core T5240 to the Sage project, but whilst its 128
threads will be OK to run a Sage server for public access, it is very difficult
to develop Sage on it, due to the slow single-threaded performance.
Given Sun are supporting the Sage project, and clearly have some interest, (see
some info at)
http://www.sun.com/customers/servers/univ_washington.xml
is there anyone reading this in Sun (or elsewhere) that can get me an account on
fast SPARC hardware?
Anyway, after asking for some help, here is an offer!!
If anyone is a *serious* developer of any software used in Sage - see list at
http://www.sagemath.org/links-components.html
or which we use to build Sage (gnu tar, gnu make, gcc, autoconf etc), then they
can get a free account on the T5240.
However, be aware, the T5240's single threaded performance is not very good, so
it is not an ideal development platform, which is why I wish to find something
better suited. But if you want to test your software on a 16-core SPARC, you can
do it for free.
Dr. David Kirkby
david DOT kirkby ATTTTT onetel DOT net.
--
I respectfully request that this message is not archived by companies as
unscrupulous as 'Experts Exchange' . In case you are unaware,
'Experts Exchange' take questions posted on the web and try to find
idiots stupid enough to pay for the answers, which were posted freely
by others. They are leeches.
Ian Collins
> Following my other question about buying cheap SPARC hardware, I might
> as well ask - there is no harm in asking.
>
> Is there anyone that can give me a *free* shell account on *fast* SPARC
> hardware to support a major open-source project
>
> http://www.sagemath.org/
>
> I've already got a Blade 2000 with 2 x 1.2 GHz, so it needs to be
> quicker than that to be any use.
>
> Sun very kindly donated a 16-core T5240 to the Sage project, but whilst
> its 128 threads will be OK to run a Sage server for public access, it is
> very difficult to develop Sage on it, due to the slow single-threaded
> performance.
Do you run parallel builds on it? Doing so will help overcome the slow
single-threaded performance.
--
Ian Collins
Greg Mortensen
>Do you run parallel builds on it? Doing so will help overcome the slow
>single-threaded performance.
This. For example, "make -j" and GCC builds are very fast on my Niagara
boxes -- much more so than on my SPARC VI machines.
--
\|/ ___ \|/ thev...@pobox.com +----- 2048R/38BD6CAB -----+
@~./'O o`\.~@ | 02BD EF81 91B3 1B33 64C2 |
/__( \___/ )__\ | 3247 6722 7006 38BD 6CAB |
`\__`U_/' +--------------------------+
Dave
Generally no, as certain packages in Sage are known to break with parallel
builds. Some parts are ok, but others break. Sage consists of about 100
packages, some of which depend on others. So sorting out what can safely be
built in parallel is not easy.
I have tended to bypass the normal method, and build ATLAS in parallel with
other components, as ATLAS takes about 12 hours to build.
The issue with ATLAS is not helped by the fact that it has no tuning parameters
for sun4v machines, so goes though a lengthy tuning process. That is on my
'todo' list. But even if that is solved, the source will take well over a day to
build on the T5240. It's a shame, as its a nice bit of hardware that is not
really being used fully.
The intension in the longer term is to have the T5240 run as a Sage server,
where many people connect to it simultaneously over the web with a browser. In
that case, the hardware will be much better utilised. But I need to sort out
some issues in the build process. I can quite easily build Sage myself, but I
want to make it easier for others to build. To do that, certain issues in the
build process need to be resolved. Doing it on a slow machine is no fun.
Dave
Dave
> In <7mtdp5F...@mid.individual.net> Ian Collins <ian-...@hotmail.com> writes:
>
>> Do you run parallel builds on it? Doing so will help overcome the slow
>> single-threaded performance.
>
> This. For example, "make -j" and GCC builds are very fast on my Niagara
> boxes -- much more so than on my SPARC VI machines.
Yes, agreed, but some packages will not build like that. It screws up the build
process. Where they are known about, 'make' can be unset. But there is about 100
separate packages, many of which depend on others.
One approach some have taken it to build in parallel, then when it fails, start
again building in parallel. Apparently this can be made to work, but it is
clearly risky. I've tended to play it safer, and not do parallel builds.