[comp.arch] Disk Striping on CM

eugene@pioneer.arpa (Eugene Miya N.) (08/05/87)

In article <214@vianet.UUCP> devine@vianet.UUCP (Bob Devine) writes:
>  Hillis' Connection Machines supposedly[*] have an implementation of
>disk striping where there are 39 disks that each get one bit from
>a word.  The word is formed by 32 data bits and 7 ECC bits.  Doing it
>in this fashion allows complete recovery even if one disk gets hosed.

The cabinet of that box appeared at the Santa Clara SuperC Convention in
May.  I was informed it was not yet hooked up.  Technically this is not
quite perceived as disk striping because its object is to have as many
processing elements/disk combinations as possible. Striped systems
are largely single CPU systems.  The fact that some have four processors
apparently deceives some users into believing they automatically get 4
CPUs working on their problems.  Sorry, automatic parallelism is not
quite all here yet (see the FX/x architectures for things which are
partially automatic).  Anyway, there is no term for what CM does.  It
was done on the ILLIAC.  The database machine people might call it a
"disk per PE" machine, but it was around before them, so I don't think
there is any standard terminology.  Suggestions? (Appeal to the
community on what you call it.)  Also, I think FPS T-Series Hypercubes
are supposed to have the capability of 1 disk per node, but I have not
seen one of thse yet, and we have an Intel.

From the Rock of Ages Home for Retired Hackers:

--eugene miya
  NASA Ames Research Center
  eugene@ames-aurora.ARPA
  "You trust the `reply' command with all those different mailers out there?"
  "Send mail, avoid follow-ups.  If enough, I'll summarize."
  {hplabs,hao,ihnp4,decwrl,allegra,tektronix,menlo70}!ames!aurora!eugene

andy@batcomputer.tn.cornell.edu (Andy Pfiffer) (08/05/87)

In article <2438@ames.arpa> eugene@pioneer.UUCP (Eugene Miya N.) writes:
>Also, I think FPS T-Series Hypercubes
>are supposed to have the capability of 1 disk per node, but I have not
>seen one of thse yet, and we have an Intel.

>--eugene miya

Actually, it is 1 disk (small SCSI) per 8 nodes.

The smallest divisable unit of a T-Series (a T-10) consists of 8 vector
boards (INMOS T414's with Weitek VPU's) and 1 system board (INMOS T414,
no VPU, 2 UARTs and a disk)

All nodes have 1 MB of memory.

...and in actuality, multiple modules (ie T-10's) are actually hooked
up as rings of 3-cubes...


	Andy Pfiffer
	Trillium Development Staff


-- 
Andy Pfiffer					andy@tcgould.tn.cornell.edu
Cornell Theory Center / Cornell U.		cornell!batcomputer!andy
Home of the first usable T-Series		(607) 255-8686
"...that's the way a Transputer works, right?"  Systems Group