Path: utzoo!mnetor!uunet!husc6!bbn!rochester!udel!gatech!sbmsg1!scbhq!uahcs1!madhat!alvitar
From: alvitar@madhat.UUCP (Phil Harbison)
Newsgroups: comp.arch
Subject: Re: Performance increase - a suggestion (really sequent cache)
Message-ID: <258@madhat.UUCP>
Date: 28 Jan 88 17:24:13 GMT
References: <235@unicom.UUCP> <8844@steinmetz.steinmetz.UUCP> <262@tmsoft.UUCP> <39245@sun.uucp>
Organization: DataVision, Huntsville AL
Lines: 28
Summary: Synapse used an ownership protocol

In article <39245@sun.uucp>, petolino%joe@Sun.COM (Joe Petolino) writes:
> > [ description of ownership protocol for cache coherence ]
> >Does anyone know if Sequent invented it (seems
> >rather too obvious for that), or a proper reference to who did? 
> 
> Amdahl was designing systems like this almost ten years ago ...

Synapse,  a  manufacturer  of fault-tolerant systems based on the 68000,
used  an  ownership  protocol on their proprietary bus.  I'm not sure if
Synapse  is still in business, but I believe they started building their
68K  systems  around 1982.  The memory controllers maintained extra bits
to  indicate  whether  any public of private copies of a memory block (4
32-bit  words)  were checked out.  Any time you performed a read and the
cache  missed, you had to do a "public" read from the memory controller.
Public  copies  of  a  memory block were read-only.  Before performing a
write,  you  had  to  do  a "private" read from the memory.  If no other
private copies were checked out, the memory controller would respond and
the  bus  watchers  on  all  other  caches would invalidate their public
copies.   If  a  private  copy  was  already  checked  out,  the  memory
controller  would  respond with an abort-and-retry signal.  Then the bus
watcher  with  the  private copy would update main memory and invalidate
its copy.

-- 
Live: Phil Harbison
USPS: 3409 Grassfort Drive, Huntsville, AL 35805-5421
UUCP: {clyde,uunet}!madhat!alvitar
PSTN: 205-881-4317