[TriLUG] Questions about Threading

[Adrian Likins]

On Wed, Mar 27, 2002 at 07:49:01AM -0500, John F Davis wrote:
> Hello
> 
> I stand corrected on the number of intel cpu's in a box.  After I saw the
> bit about 16 cpu's, I went and asked one of my colleagues.  He said he had
> heard of boxes supporting more than 4 but he didn't know the upper limit.
> He said 8 or 16 maybe the upper limit.  However, what is the upper limit
> for Sun boxes? I'm pretty sure they shipped some 256 cpu monsters and that
> is the point about their more complex (slower)  threads.  They are slower
> (compared to linux)  on boxes with a few cpu's, but with many cpu's they
> scream.
> 
> JD
> 
> Thanks for the odslab url.

	Well, seems like sequent^H^H^H IBM numaQ boxes
are good for 64 pIII's (and I've heard rumours of SCO
actaully running on that hardware...). 

	SGI claims the Origin 2000 series can go
to 512 processors (http://www.sgi.com/origin/2000/).
I'm no big hardware expert, but I belive those fall
somewhere between a "mainframe" and a cluster. I
belive those are based on the same design ideas
as the sun e10k's (iirc, these were based on
a cray design that was bought by SGI and then
sold to sun). But I could easily be wrong...

	hp superdomes seem to fit into about
the same range as e10k's and friends.
http://www.hp.com/products1/servers/scalableservers/superdome/specifications.html

	It seems like at one point Tandem
had 64way mips based machines as well, but that
may be ancient history.

	The fujistu prime power2000 claims 128
cpu smp support. It appears to be a big honking
sparc/solaris box. Wouldnt surpise me if it's
the same arch as the e15k's
http://primepower.fujitsu.com/en/m2000.html
http://www.tpc.org/tpcc/results/tpcc_result_detail.asp?id=101082801

	unisys and NCR/Teradata both have 64-128
processor x86 based machines as well. The Teradata
worldmark.

btw, a great place to browse for big hardware is
the tpc benchmark results, see http://www.tpc.org


Adrian