home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!cis.ohio-state.edu!pacific.mps.ohio-state.edu!linac!unixhub!kinetics.slac.stanford.edu!user
- From: dbg@slacvm.slac.stanford.edu (David Gustavson)
- Newsgroups: comp.arch
- Subject: Re: BUSES
- Message-ID: <dbg-270792195520@kinetics.slac.stanford.edu>
- Date: 28 Jul 92 02:59:29 GMT
- References: <1992Jul23.092211.18462@nuscc.nus.sg> <1992Jul23.191927.1181@pcnntp.apple.com> <1992Jul27.191347.4485@ksmith.uucp>
- Sender: news@unixhub.SLAC.Stanford.EDU
- Followup-To: comp.arch
- Organization: SLAC
- Lines: 33
- Nntp-Posting-Host: 134.79.17.25
-
- In article <1992Jul27.191347.4485@ksmith.uucp>, keith@ksmith.uucp (Keith
- Smith) wrote:
- >
- > IMHO the only way to solve this is to have seperate input and output
- > "busses" that allow for multiple electrons to be traveling across the
- > "wires" at the same time. Seperate because you can be sending fetch
- > requests off of one side while yanking them in on the other.
- >
- Well, not electrons exactly, but signals. You're right that buses are
- a bottleneck, because devices have to take turns using them, and the
- solution to that requires lots of independent paths (at least two for
- each processor) that can signal at the same time if you don't want the
- processors to be waiting needlessly for access to the bus.
-
- The latest 'bus' uses that approach, and is called the Scalable Coherent
- Interface. It recently became approved as IEEE Std 1596-1992. It will
- first be used in super multicomputers, but soon in some interesting
- workstation-class machines too.
-
- SCI uses fast narrow unidirectional differential links that run at
- 1 GByte/s for short distances (meters), and slower bit-serial fiber
- optic (up to 10km) or coaxial (up to 20m) links at 1 Gbit/s.
-
- SCI also provides a scalable cache-coherence mechanism to support
- distributed shared memory systems, and a variety of multiprocessor
- supporting features (queue reservations, fair access, forward progress
- guarantees).
- --------------------------------------------------------------
- -- David B. Gustavson, Computation Research Group, SLAC, POB 4349 MS 88,
- Stanford, CA 94309 tel (415)926-2863 fax (415)961-3530
- -- What the world needs next is a Scalable Coherent Interface!
- -- Any opinions expressed are mine and not necessarily those
- of the Stanford Linear Accelerator Center, the University, or the DOE.
-