[comp.os.research] Technical Report

carla@macbeth.cs.duke.edu (Carla Ellis) (06/01/90)

We would like to announce the availability of the following technical
report that describes measurements taken with our DUnX kernel comparing
various page migration and replication policies:


  Experimental Comparison of Memory Management Policies for
                 NUMA Multiprocessors

   Richard P. LaRowe Jr. and Carla Schlatter Ellis

            Department of Computer Science
                   Duke University
                  Durham, NC  27706

                      April 1990

                      CS-1990-10

                       ABSTRACT

       Non-uniformity of memory access is an  almost
  inevitable  feature  of the memory architecture in
  shared  memory  multiprocessor  designs  that  can
  scale  to large numbers of processors.  One impli-
  cation of NUMA architectures is that the placement
  and  movement  of  code and data become crucial to
  performance.  As memory architectures become  more
  complex  and  the non-uniformity becomes less well
  hidden, system software must assume a larger  role
  in  providing  memory  management  support for the
  programmer.  This paper investigates the  role  of
  the operating system.

       We take an experimental approach to  evaluat-
  ing  a  wide-range  of memory management policies.
  The target NUMA environment is BBN's GP-1000  mul-
  tiprocessor.   Extensive  local modifications have
  been made to the memory  management  subsystem  of
  BBN's nX operating system to support multiple pol-
  icy implementations.  Policy comparisons are based
  on  the  measured  performance  of  real  parallel
  applications.

       Our  results  show  that  there  are   memory
  management policies implemented in our system that
  can improve the performance  of  programs  written
  using the simpler uniform memory access (UMA) pro-
  gramming model.  While achieving the level of per-
  formance of a highly tuned NUMA program is still a
  difficult  problem,  some  examples  come   close.
  There  appears  to be no single policy that can be
  considered the best over our set of test  applica-
  tions.  Investigations into the contributions made
  by  individual  policy  features  toward   overall
  behavior  of  the  workload  provides some insight
  into the design of a set of effective policies.




           DETAILS ON HOW TO ACQUIRE REPORT

The technical report described above is available for
anonymous ftp from cs.duke.edu (128.109.140.1)

It is a compressed postscript file in /pub
called DUnX-TR.ps.Z

To get a hardcopy version via USmail send a request for
TR CS-1990-10 to
carla@cs.duke.edu or rpl@cs.duke.edu.

Also please send any comments or questions you have
after reading the report.