commit 4039eab5220586126f8e4878b1b423092e7764b1
parent c96ea7fb3f162181a8584f250d4f01dd4b15f5b5
Author: Ivan Gankevich <igankevich@ya.ru>
Date: Mon, 15 May 2017 16:59:38 +0300
Rewrite scenario 2 and 3.
Diffstat:
1 file changed, 9 insertions(+), 5 deletions(-)
diff --git a/src/body.tex b/src/body.tex
@@ -223,7 +223,12 @@ the first surviving node from the IP address list of neighbours. If such node
is online, the search stops and the subordinate is deleted. If the node is not
found, the subordinate restores the principal from the copy on the current node
and deletes itself. This algorithm is executed on every node, to which a copy
-of the principal was sent. Subordinate deletion is necessary, because the whole
+of the principal was sent, and the guarantee that only one copy of the
+principal is restored is provided the implied hierarchy of IP addresses: every
+subordinate of the principal has the list of nodes to which only
+\emph{previously created} subordinates were sent, and communication originating
+from previously created subordinate to the newer subordinate is possible (only
+the other way round). Subordinate deletion is necessary, because the whole
computational step, modelled by the principal, is re-executed from the initial
state, and there is no simple and reliable way of taking into account partial
results which were produced so far by the subordinates.
@@ -307,10 +312,9 @@ involves the use of persistent storage and there is no reliable way to ensure
that all cluster nodes have been bootstrapped. If some nodes were not
bootstrapped properly, missing kernels are considered failed in accordance with
the first and the second scenarios. This may lead to re-execution of
-considerable portion of parallel programme kernels, especially when multiple
-principal kernels in the same hierarchy branch have failed. If a node fails in
-the middle of recovery process, the whole process is restarted from the
-beginning.
+considerable portion of parallel programme, especially when multiple principal
+kernels in the same hierarchy branch have failed. If a node fails in the middle
+of recovery process, the whole process is restarted from the beginning.
%\begin{figure}
% \noindent%
