🚀 go-pugleaf

 A colleague of mine is conducting research on placement strategies for
 objects in distributed object-oriented systems. His work involves
 simulating object interaction and migration and, rather than using
 totally artificial numbers, he is seeking realistic statistical
 distributions representing the characteristics of object-oriented
 systems. He would be interested to learn of any source of such
 statistics. I append a statement of what he is seeking, expressed in
 his own words lest I misrepresent his requirements.

 Cheers....Ron

                   ---------------------------------

As part of my research, I need to simulate a network of computers
which support object migration. Hence I either need to use an existing
simulator (after modification of course) or build one. Are any
simulators available?

In order to drive the simulator and make the simulation as realistic
as possible, I would like some real world data (if it exists)
concerning object populations.

For example, I would like to know:-

  what is the typical population density of objects on a node?

  what is the distribution of sizes (in bytes) of those objects?

  how many methods does the typical object provide, and how much memory
  do these methods consume?

  what sort of interactions do the objects undertake?
    the typical number of clients per server?
    how many servers does a typical client use?
    how many 'references' can a typical object store (ie what is the
    potential for relationships or connections between objects)?
    (Those who have written reall applications should have some feel
    for this one.)

  how frequently is the typical object invoked (accessed)?
    what is the ratio of 'read' to 'write' accesses?

  how long does the typical object live?

In the absence of any data from real OO systems, I could look at the
'objects' stored by file systems, or database 'objects', and the
'objects' managed in process migration systems, but is anyone aware of
similar studies for *real* objects?  Has anyone recorded traces of OO
programs in execution; are the traces available?

Alternatively, has anyone instrumented and traced, say, NFS to record
remote file access patterns?

Has anyone instrumented a garbage collector to record the sizes of the
objects recovered?

In systems which support load sharing (either initial-placement or
process migration), how much is actually moved between machines (ie
what size is the process space)?

-- Jonathan

===========================================================================
>From  : Jonathan M Spencer
Mail  : Computing Lab., University of Newcastle-upon-Tyne, NE1 7RU, England
Phone : +44 91 222 8229
ARPA  : J.M.Spencer%newcastle.ac.uk@cs.ucl.ac.uk
JANET : J.M.Spencer@uk.ac.newcastle
UUCP  : !uknet!newcastle.ac.uk!J.M.Spencer
===========================================================================