Paper #:[2.2 WP] #8
Title: "Managing Update Conflicts in Bayou, a Weakly Connected Replicated Storage
System" by Douglas B. Terry, Marvin M. Theimer, Karin Petersen, Alan J. Demers, Mike
J. Spreitzer and Carl H. Hauser

PROBLEM
Mobile devices tend to lack continuous connectivity to the network.  Consequently in
order to access data without a connection, the user will utilize a local copy of the
data on his device for reading and editing purposes.  Unfortunately, after he edit
the document, someone else may have been in the same situation.  Consequently, when
both of the edited documents are uploaded to the server, a conflict occurs. Bayou is
a solution to try to minimize and resolve the update conflicts.  However, instead of
looking at the raw data, Bayou uses other available information, such as application
semantics in order to help with change interpretation.  The authors focus on a
meeting room scheduler and a bibliographic database in order to show the usefulness
of the Bayou approach, which consists of servers using anti-entropy sessions to
converge to a consistent state.

NEW IDEAS AND STRENGTHS
*Utilizing the semantic information known by applications about what data may
conflict and consequently, creating a plan at the application level to handle the
conflict.  For example, the room scheduler has the user select multiple times in
order to handle conflicts and if none of the desired times are available, the
application selects the closest available time.
*User does not have to wait for the update to "propagate" to all of the servers in
order to have access to a file.  The researchers assume that the user will not
object to a few conflicts in order to ensure having access to the file at all times.
*While anti-entropy occurs in a different order for different servers, by using the
writeID the servers eventually converge to the same contents, while at the same time
the merges only depend on each server's current data.
*It was a good idea to use public-key cryptography for security purposes to help
secure the data in the data collections.

WEAKNESSES AND EXTENSIONS
*The researcher's implementation has to store the database into virtual memory,
which limits scalability.  However, on the bright side, this is not a requirement of
Bayou.
*Bayou uses the primary commit scheme which means that the primary server is
responsible for committing updates.  However, this could be a weakness if the
primary server is disconnected from the network for an extended period of time.
This is something that could use an extension to improve.