CPSC 449: Honours Thesis
Joanna Knowitall
12352881
knowitall@cs.ubc.ca
September 28, 1992

Supervisor: Jim Blah
Project: Neverland

	
        Neverland   is a distributed file system,    dealing with storage of
files across a system of distributed machines. It maintains a certain number
of copies of each file on the client across  all the connected machines. For
each newly   created  file on the  client   machine,  a policy  is  created,
determining  such things  as  the   number  of replicas  necessary, and  the
frequency  of updates. This is presented  in the form of  a  table that maps
each node to these values.

        In addition  to  the   client   machine,  there  is   a  network  of
interconnected  replica nodes. These monitor  each other constantly, to pass
on the updated copies of   files and also to  detect  failure of any of  the
nodes. One of these nodes, usually the one  that has the smallest 'staleness
factor'   - in this case, this   term simply means how often   a file on the
client machine is replicated  to this node -  is chosen  as the leader.  The
leader's  difference only comes into  play when one of  the  nodes fails. In
that case, all the other nodes inform the leader of  the failure. It is then
the leader's job to choose a new replica node to  replace the failed one. It
then builds a   list/table  of files that   were  on the  failed   node, and
propagates the  list to the new  node. From this point,   the new node takes
over, and the  leader goes back  to its  regular  job. The new node  must go
through the list provided, and obtain copies of all the files. Since this is
the part my project will be dealing with, specifically, I will discuss it in
more detail below.

        The problem   I  will be  dealing   with can be  summarised   as the
following:  given a set  (list) of files, how  to efficiently pull the files
from a  number of replica  nodes  spread out across  a  network. A number of
factors must be   considered. One requirement  is  that the new  node should
probably get the newest  copies of each file  available, but not if  that is
likely  to  otherwise  compromise the  efficiency   of  the  process.  Other
considerations are  bandwidth available  and   load on the nodes,  both  the
sending   and receiving node,  which would  limit  your  ability to copy the
files,   and make  an   otherwise  attractive  node  an inefficient  choice. 
Connection limits factor into the decision, as they would also influence the
time it would take to fully populate the new node  with the necessary files. 
However, these factors  should probably be secondary  to the currency of the
file version, and used to break ties and the like.

        The rough outline of the work plan is as follows.  Over the next few
months, I will be looking at various publications in the field over the past
few years, and   getting acquainted with the  system  (and code)  that is in
place. I have a few sources that I can start with, that have been referenced
in previous research on Neverland, and from there I should be able to branch
out into different sources. A  good area to   look in would be  peer-to-peer
systems, and those dealing with storage, in particular, as well as searching
for articles on  replication, how it's done  and what are the most efficient
ways to do  it. The research on replication  and file systems should help me
get a better grasp on the area, and also compare this system against similar
ones  that may have been  developed. This will help  me figure out where the
particular piece that I will be working on can fit best.

        By  the end of  November I should have  completed  a design document
outlining the structure and functionality of  the proposed component. Actual
implementation should begin soon after  that, going into second term. Having
done the  research  on similar file   systems with similar  functionality, I
should have  enough information to be able  to compare  my implementation to
others that have  been proposed or  finished. The implementation and testing
should be done  by March, and the rest  of the time  will be spent compiling
the results and writing the report.