home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.parallel
- Path: sparky!uunet!paladin.american.edu!gatech!hubcap!fpst
- From: dbg@SLAC.STANFORD.EDU (David B. Gustavson)
- Subject: Re: Bus arbitrator and lock implementation
- Message-ID: <1992Dec31.140910.3848@hubcap.clemson.edu>
- Apparently-To: <comp-parallel@uunet.uu.net>
- Followup-To: comp.parallel,comp.unix.programmer,comp.sys.intel.
- Sender: news@UNIXHUB.SLAC.STANFORD.EDU
- Organization: SLAC Computation Research Group
- References: <1992Dec28.160326.27290@hubcap.clemson.edu>
- Date: Thu, 31 Dec 1992 05:30:12 GMT
- Approved: parallel@hubcap.clemson.edu
- Lines: 40
-
- In article <1992Dec28.160326.27290@hubcap.clemson.edu>,
- sundaram@eng.uiowa.edu (Subbiah Sundaram) wrote:
- > ... (In particular I would like to know about
- > the bus arbitrator implementations on shared memory parallel
- > machines.(Books, papers, articles, etc.))
- >
- It depends how parallel a machine you want. For small parallel machines,
- where you can tolerate a bus backplane/bottleneck, you could look at IEEE
- Std 896.1, Futurebus+, and 896.2. Futurebus+ has two arbiter schemes, one a
- distributed bused one with many bits of priority for realtime scheduling
- applications, and a faster one using centralized arbiter logic.
-
- If you want a shared-memory parallel machine that can scale up to larger
- applications, you should take a look at ANSI/IEEE Std 1596-1992, the
- Scalable Coherent Interface. The concept of arbitration is quite different
- there, because it was designed for a high degree of parallelism. It only
- acts like a bus, providing bus-like services to the user--actually it uses
- many independent links that can be active at the same time, performing
- independent communications. SCI also defines multiprocessor synchronization
- operations and locks that can be exported through an arbitrary interconnect
- (ring, switch, mesh, whatever)--they don't use the strategy of bringing the
- whole system to a momentary halt like the old bus-based locks did (using
- read-modify-write operations and the bus bottleneck to best advantage).
- However, if you're in a hurry, keep in mind that SCI isn't really available
- yet (the SCI interface chip, which includes all the transceivers, FIFOs,
- and protocol logic has slipped to Jan/Feb for the GaAs version, which
- supports two 1-gigabyte/s links; the CMOS version isn't expected until
- March, and will be slower but cheaper).
-
- You can get the documentation on Futurebus+ or SCI from the IEEE the same
- way you get any other IEEE standard. Let me know if you want more info on
- SCI, I have a few Email status files I can mail you.
-
- --------------------------------------------------------------
- -- David B. Gustavson, Computation Research Group, SLAC, POB 4349 MS 88,
- Stanford, CA 94309 tel (415)961-3539 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.
-
-