Re: dumb question about multi-core processors...
- From: Greg Folkert <greg@xxxxxxxxxxxxxxx>
- Date: Fri, 16 Feb 2007 18:57:51 -0500
On Fri, 2007-02-16 at 16:37 -0600, Ron Johnson wrote:
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On 02/16/07 15:58, Michael Fothergill wrote:
Dear Debianists,
I read an article:
http://www.newsfactor.com/story.xhtml?story_id=032003A1JK8W
It says that Intel has made a chip with 80 processors on it that it says
is a teraflops device and only uses a low power consumption.
But the article suggests that it is tough to write an OS to take
advantage of a large number of such processors.....
Dumb question 1:
How tough would it be to produce a version of Debian that could run this
sort of chip if a commercial version of it were made available?
The existing x86-64 kernel should handle an 80-core system.
Umm, maybe *AN* 80-core processor, but not that particular one. That
particular one was an 80-core VLIW processor or a Very Long Instruction
Word Processor. Linux has already been developed to handle such
processors. It is just a matter of knowing the Instruction set and
getting the compilers and libraries to conform. Which in all reality, as
of now it a non-issue. Linux may just well be the first OS Kernel to
even run on such a processor.
The problem with high-CPU-count SMP systems is coordinating the
processes, sharing cache and RAM. NUMA was created to mitigate this.
Well, a VLIW process already would have been designed to handle most of
these issues... NUMA was indeed made to handle this stuff, but once
again VLIW processors are slightly different. Sort of like the
difference between PA-RISC or SPARC and x86*
Depending on the physical design of the CPU & glue logic, and how
how the OS is written, adding CPUs it might show unacceptably low
incremental improvement after 2, 4, 8, etc CPUs. SparcSolaris can
handle up to 128 CPUs on their big system.
And Linux actually for a short time, supported that behemoth better than
Solaris did. If only for a few weeks.
Obviously the the application mixture also affects the suitability
of N-way SMP systems.
Indeed, nothing like having a a HUGE POWERFUL Multi-Million Barrel Oil
Tanker to transport a pallet of Water Bottles atop its deck and nothing
else.
Dumb question 2:
Could it be that if Intel puts out a chip or array of chips or whatever
its called of this type that the existing OS companies like MIcrosoft
and Apple would take so long to figure out and implement an OS that run
on it that Linux might be the only usable option for such a long time
that it might even end up with an effective monopoly here?
Fat chance. People also said that the expense of upgrading to Vista
would incite a great migration to Linux. Hasn't happened. Won't
happen.
Well, we have yet to really see. But I suspect once the final cost and
dissatisfaction it produces, we might see some opportunity to hook
some(much?) of that angst.
--
greg@xxxxxxxxxxxxxxx
Novell's Directory Services is a competitive product to Microsoft's
Active Directory in much the same way that the Saturn V is a competitive
product to those dinky little model rockets that kids light off down at
the playfield. -- Thane Walkup
Attachment:
signature.asc
Description: This is a digitally signed message part
- Follow-Ups:
- Re: dumb question about multi-core processors...
- From: Ron Johnson
- Re: dumb question about multi-core processors...
- References:
- dumb question about multi-core processors...
- From: Michael Fothergill
- Re: dumb question about multi-core processors...
- From: Ron Johnson
- dumb question about multi-core processors...
- Prev by Date: Re: Woody on 486 problem
- Next by Date: Re: Woody on 486 problem
- Previous by thread: Re: dumb question about multi-core processors...
- Next by thread: Re: dumb question about multi-core processors...
- Index(es):
Relevant Pages
|