|
|
Software and hardware
failures may lead to loss of valuable data. Replication, is
indinsbensable in ensuring data survaivability and availability. Data
is replicated among multiple basic storage units - servers or disks.
Although replication copes with loss of data, it fails to provide
consistency guarantees when multiple processes access
different replicas concurrently. To assess the state of the replicated
data, researchers proposed various levels of consistency semantics.
Atomicity is the strongest and most precise consistency semantic, and
therefore the most difficult to supply in a distributed, asynchronous,
failure prone setting. Several solutions proposed atomic read or write
storage implementations. However, the inherent complexity of atomicity
led these approaches to incorporate excessive communication between the
clients and the data replicas, deteriorating the operation latency of
every
read or write operation.
This project investigates the operation latency of read and write
operations of atomic register implementations that support multiple
writers and multiple readers (MWMR). Such implementations comprise
building blocks for more complex Distributed Storage Systems (DSS).
Also, they can be used directly to implement distributed file systems,
which are interesting on their own right. In summary this work aims to
investigate:
(a) Is it possiple to devise algorithms implementing MWMR atomic registers with low operation latency?
(b) What is the practicality of such algorithms in the context of experimental simulations and
implementations in planetary scale networks? |
|