[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Broken applications: Could we be honest?

On 7/8/06, Art Edwards <edwardsa@afrl.kirtland.af.mil> wrote:
I have been writing to the list about two applications that
are so broken on the AMD64 distribution that they render the
box pretty useless. I'm sure one could say that two measly
applications are no big deal. However, if you do scientific computation
for a living, and two of the primary tools are broken, you now have
a rather clumsy paperweight where a computer should be. You could
argue that we should simply learn new tools, and we could, but we
should really be doing science instead.

This brought up the question, who uses 64 bit Linux anyway?
Surely gamers do not drive the 64-bit linux community. It can't be the desktop
community, seeing that the standard office tool doesn't really
work for 64-bit. I would think that scientific and engineering
users would drive this community. Besides the instruction set, which
can probably give some speed, but wouldn't justify the cost, the address
space in 64-bit OS's mean that we can solve much larger problems. Unles you're
not doing some heavy-duty, memory-intensive computation, 64-bits seems
to be simply a status symbol.

For compute servers the amd64 distribution is fine. All you really need are
languages (compilers), libraries and decent MPICH. We run our small, 32 bit
Beowulf on debian with abandon, and from my experience, I look forward to
converting it to amd64 ... with a 32-bit node where things actually work.

Unless such core pieces as the debugging tool (ddd) and the data display tool
(xmgrace) are working, it is dishonest to pretend that the 64-bit version
is ready for testing. It would be very nice if you, and other distro's, were
to put appropriate caveats on the websites, saying that 64-bit is really not
ready for the prime-time desktop. That way, we could make better purchasing

At the risk of imposing what we do at our work place onto your work flow, i find that users generally should have access to better debuggers/profilers than what ships with standard gnu distros. presumably if you are doing scientific computations, you probably have access to a commercial compiler? i know that the portland group compilers ship with a fairly good gui debugger if you are not satisfied with gdb (in parallel attached to each running process)

also shouldnt users be using programs like xmgrace on their local workstations? again with out trying to impose my workflow to yours, i find sometimes users do silly things on the head node on clusters, and I tend to try and get my users to do post analysis etc... stuff that can run serially on their own desktops whenever possible.


Jimmy Tang
Trinity Centre for High Performance Computing,
Lloyd Building, Trinity College Dublin.
Reply to: