Author: Ivan Gankevich <firstname.lastname@example.org>
Date: Fri, 24 Mar 2017 20:53:14 +0300
1 file changed, 13 insertions(+), 0 deletions(-)
diff --git a/src/tail.tex b/src/tail.tex
@@ -51,7 +51,20 @@ kernels upon a failure.
+In the paper we propose a system architecture consisting of two tree
+hierarchies of entitites, mapped on each other, that simplifies provision of
+resilience to failures for parallel programmes. The resilience is solely
+provided by the use of hierarchical dependencies between entitites, and is
+independent on each layer of the system. To optimise handling failure of
+multiple cluster nodes, we use the hierarchy implied by the order of creation
+of subordinate entitities. The hierarchical approach to fault tolerance is
+efficient, scales to a large number of cluster nodes, and requires slow I/O
+operations only for the most disastrous scenario~--- simultaneous failure of
+all cluster nodes.
+The future work is to standardase application programming interface of the
+system and investigate load-balancing techniques, which are optimal for a
+programme composed of many computational kernels.
The research was carried out using computational resources of Resource Centre