Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!csd4.milw.wisc.edu!nic.MR.NET!srcsip!haarlem!shankar
From: shankar@haarlem.SRC.Honeywell.COM (Son of Knuth)
Newsgroups: comp.arch
Subject: Re: Content Addressible Memories
Message-ID: <15965@srcsip.UUCP>
Date: 2 Feb 89 00:15:50 GMT
References: <1036@cmx.npac.syr.edu> <28200263@mcdurb>
Sender: news@src.honeywell.COM
Reply-To: shankar@haarlem.UUCP (Son of Knuth)
Organization: Honeywell Systems & Research Center, Camden, MN
Lines: 25

In article <28200263@mcdurb> aglew@mcdurb.Urbana.Gould.COM writes:
>
>>/* Written  9:22 pm  Jan 25, 1989 by hamid@goedel.uucp in mcdurb:comp.arch */
>>I recently developed a Prolog Abstract Machine for content-addressable
>>memory (along the lines of the Warren Abstract Machine).  This new abstract
>>machine is shown to be superior to the WAM in many respects. It is discussed
>
>So, how much CAM do you require for your PAM?
>All of memory? Or can you have just a small bit of CAM and a large
>amount of regular RAM?

Well, perfect opportunity to give a plug for my paper presented at last
year's ICLP (Intl Conf Logic Programming).  I essentially used
a small amount of CAM and larger amounts of RAM arranged in a hierarchy,
and my performance estimates seemed to show that you could get good
performances through such an approach due to the locality inherent
in Prolog.  That approach only used CAM to store rules for unification
rather then for binding environments, though.

In short, the answer is yes, you *can get performance benefits without
needing huge amount of CAM storage.
---
Subash Shankar            Honeywell Systems & Research Center
voice: (612) 782 7558     US Snail: 3660 Technology Dr., Minneapolis, MN 55418
shankar@srcsip.uucp    {umn-cs,ems,bthpyd}!srcsip!shankar