Linux distributives for Opteron
Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.
Donald Becker becker at scyld.comThu Mar 20 13:42:13 PST 2003
- Previous message: Linux distributives for Opteron
- Next message: CCL:Multiprocessor scheduling G98/Linda on a Linux cluster (fwd)
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
On Thu, 20 Mar 103, Mikhail Kuzminsky wrote:
> Subject: Linux distributives for Opteron
>
> By RedHat it was declared that x86-64 support will be realized
> in the frames of RH Linux Advanced Server distributive.
> But what is known about much more cheap RH Linux professional (according
> our experience, it is enough for building beowulf clusters w/2-cpu's
> nodes) ? According my data, x86-64 support in RH may be realized
> just at summer.
As I have hinted in my postings (and at our SC02 demo), we have been
working on Opteron support. Although I don't see any race to be first
with a no-cost, full-support release.
The x86-64 backwards-compatible approach means that there are several
different levels that can be supported:
Fixing the few minor issues with running entirely in 32 bit mode.
(Easily done. Our distribution and I believe most others have already
made the changes.)
A release with a 64 bit kernel, both 32 and 64 bit libraries and
compiler chain, using 32 bit utilities and applications. This
involves tracking down some obscure library bugs(*), but most
everything works. With a few days work you can do this at home.
It took a few weeks of work for our system.
A 64 bit kernel, with a mix of tools compiled for 32 and 64 bits as
needed.
A 64 bit release, with all tools recompiled and 32 bit libraries only
used by third party applications.
A release with a 64 bit kernel, dynamically supporting both 32 and 64
bit compute nodes. That means
detecting and providing the correct kernel to the compute nodes,
limiting x86-64 applications to nodes that can support them
controlling application processes on either 32 or 64 compute nodes
*1 How can 'sleep' dump core? Oh, when it's GNU sleep that has support for
converting from extra-planetary time units and then calling
nanosleep(). Grrrr...
--
Donald Becker becker at scyld.com
Scyld Computing Corporation http://www.scyld.com
410 Severn Ave. Suite 210 Scyld Beowulf cluster system
Annapolis MD 21403 410-990-9993
- Previous message: Linux distributives for Opteron
- Next message: CCL:Multiprocessor scheduling G98/Linda on a Linux cluster (fwd)
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
