[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[hagmann.pa: Announcement of April 10th BASS]



note Kemph's talk....

     ----- Begin Forwarded Messages -----

Date: Mon, 30 Mar 87 08:48:05 PST
From: hagmann.pa
Subject: Announcement of April 10th BASS
To: ComputerResearch^.pa
Reply-to: hagmann

Below is the BASS announcement from the SUN coordinator for BASS, David
Rosenthal.  Please respond to me by electronic mail by Monday, April 6
if you plan to attend.  There is a total of 80 person limit in the
conference room, so attendance may be limited (sigh).  If limited, the
order of response to this message will determine priority.  The map
below looks best in fixed pitch font.  The location is the old Xerox
Geng Road site.





              BASS 8:  Databases & Other Stuff

                    Friday,  April 10th



                      Sun Microsystems
                       2300 Geng Rd.
                         Palo Alto





1.  Location

BASS 8 will be held in the Glacier Meeting Room at  the  new
location  for  Sun's  Software Products Division,  2300 Geng
Rd.,  Palo Alto on Friday April 10th.  X marks the spot.


                 Oregon  |   | |  Embarcadero      North(ish)->
                         |   | |
                         |   | |
        =================+===+=+===========================
                 101     |   | |
        =================+===+=+===========================
                   \     |   | |   /
                    -----+---+-+---  ---------------Frontage
                     \  /   \| |    /         X  O
                      --     | |   /       O
              ---------------%-%---     O    Park
                             | |             |
                             %-%-------------+-------Geng
                             | |


Refreshments and lunch will be served.

2.  Program


09:00   Coffee & Pastries
09:30   NonStop SQL            Jim Gray          Tandem
10:30   Starburst              John McPherson    IBM
11:30   Coffee
11:45   Hiding Secrets         Dorothy Denning   SRI
12:45   Lunch
 1:45   Mahler                 David Wall        DEC WRL
 2:45   Tea
 3:00   Portable CommonLoops   "jak"             H-P Labs


2.1.  NonStop SQL


                      Jim Gray: Tandem


NonStop SQL is an implementation of the  ANSI  Standard  SQL
relational database language.  Extensions in the data defin-
ition area include the notions of  catalogs,  assertions  on
tables,  and table partitioning,  organization,  node auton-
omy features,  and integration  with  the  host  naming  and
authorization  mechanism.   Extensions in the data manipula-
tion are include binding of the SQL source an  dobject  with
compiled  object  programs,   location  independent  naming,
locking extensions,  and extensions to deal with both  tran-
saction protected files and non-transactional files.

The implementation goal was to provide the functionality and
useability    of   SQL   with   the   price/performance   of
IMS/Fast Path.  Benchmarks demonstrate that the  system  can
deliver  well  beyond  200  transactions per second at about
$50K/tps.

After a brief sketch of SQL and the NonStop SQL  extensions,
the  talk  will  focus on the design choices which gave such
impressive results.

2.2.  The Data Management Extension  Architecture  of  Star-
burst


       John A. McPherson, IBM Almaden Research Center


Starburst is an experimental relational database project  at
IBM  Almaden  Research Center.  The Starburst project's goal
is to produce a portable, extensible,  distributed  database
management  system  which incorporates new system structures
and performance techniques.

In support of the extensibility goal, a database  management
system  architecture  has  been defined that facilitates the
implementation of data management extensions.   This  archi-
tecture  supports two classes of data management extensions:
"flavors",  which  are  relation  storage  extensions,   and
"attachments",  which  are  access  method or integrity con-
straint extensions.  Generic sets of operations are  defined
for  flavors  and  attachments, and these operations must be
provided in order to add a new flavor or attachment type  to
the system.  The data management extension architecture also
provides common services  for  coordination  of  flavor  and
attachment execution.

This talk will  give  a  brief  overview  of  the  Starburst
project  and  then  will discuss in more detail the database
management extension architecture.

2.3.  Hiding Secrets in Databases


           Dorothy E. Denning: SRI International


SRI International and Gemini Computers are developing a for-
mal  security  model  for  a relational database system that
would protect sensitive data  of  different  classifications
(e.g.,  proprietary,  confidential, or classified in the DoD
sense).  Two principal problems that the model must  address
are:  assurance  that sensitive data of a given class cannot
be accessed by or leaked to users who are not authorized for
data of that class, and the avalanche of effects on the data
model that arise from hiding data.

The  talk  will  discuss  these  problems  and  describe  an
extended relational model that solves them.

2.4.  The Mahler Experience: Using an Intermediate  Language
as the Machine Description


                    David Wall: DEC WRL


Division of a compiler into a front end and a back end  that
communicate  via  an  intermediate  language is a well-known
technique.  In building compilers for a family  of  machines
with simple instruction sets and addressing capabilities, we
have designed an intermediate language that we  use  as  the
official  description  of  the  machine,  hiding some of the
inconvenient idiosyncrasies of the  real  machine  from  the
users and the front end compilers.

To do this credibly, we  have  had  to  hide  not  only  the
existence  of  the  idiosyncrasies  but also the performance
consequences of hiding them.  That is,  the  back  end  that
compiles  and  links the intermediate language tries to gen-
erate code  that  does  not  suffer  a  performance  penalty
because  of  the details that were hidden from the front end
compiler.  For the most part we have been successful.

In this talk we discuss the techniques  used  to  accomplish
this.   We  use  interprocedural register allocation to hide
the fact that the machine has only a finite number of regis-
ters.   We  use instruction scheduling to hide the existence
of  multi-cycle  floating-point  instructions  and  pipeline
stalls.   Finally,  we  use  branch slot filling to hide the
existence of delayed branches.


2.5.  Experiences with Portable CommonLoops


                      "jak":  H-P Labs


  -> Dan:
  ->    I expect to have some results on CommonLoops by April 1,
  -> the due date for papers submitted to OOPSLA and would like to
  -> give something at BASS also. Is this possible? The paper will
  -> be called "Experiences with Portable CommonLoops" and will
  -> contain a discussion of Warren Harris's window system, my
  -> portable CommonObjects implementation, and some profiling
  -> results. I expect the profiling to include, at least, HP Lisp,
  -> and possibly also Kyoto Common Lisp and Franz Lisp. Remotely,
  -> I might be able to get Lucid, Symbolics, and DEC Lisp in too.
  ->
  ->
  ->            jak


     ----- End Forwarded Messages -----