From 752fa10f606bb12dd8f50a1e197c18057a25e7e8 Mon Sep 17 00:00:00 2001
From: Chris Broekema <broekema@astron.nl>
Date: Wed, 2 May 2007 13:07:02 +0000
Subject: [PATCH] BugID: 1011 Add support for the "Storage.PsetsPerStorage"
option, allowing more than one Pset to write to a storage node.
---
Appl/CEP/CS1/CS1_Run/src/CS1_Sections.py | 50 +++++++++++++-----------
1 file changed, 28 insertions(+), 22 deletions(-)
diff --git a/Appl/CEP/CS1/CS1_Run/src/CS1_Sections.py b/Appl/CEP/CS1/CS1_Run/src/CS1_Sections.py
index c64abf810b4..1ba273d02e2 100644
--- a/Appl/CEP/CS1/CS1_Run/src/CS1_Sections.py
+++ b/Appl/CEP/CS1/CS1_Run/src/CS1_Sections.py
@@ -84,27 +84,30 @@ class InputSection(Section):
host = copy.deepcopy(myhost)
slaves = host.getSlaves()
- inputNodes = parset.getInputNodes()
- outputNodes = range(1, self.nCells + 1)
- allNodes = inputNodes + [node for node in outputNodes if not node in inputNodes]
-
- inputIndices = range(len(inputNodes))
- outputIndices = [allNodes.index(node) for node in outputNodes]
+ inputNodes = parset.getInputNodes()
+ outputNodes = range(1, self.nCells + 1)
+ allNodes = inputNodes + [node for node in outputNodes if not node in inputNodes]
+
+ inputIndices = range(len(inputNodes))
+ outputIndices = [allNodes.index(node) for node in outputNodes]
- newslaves = [slaves[ind - 1] for ind in allNodes]
- host.setSlaves(newslaves)
- self.noProcesses = len(newslaves)
+ print("slaves %d" % len(slaves))
+ print("allNodes %d" % len(allNodes))
+
+ newslaves = [slaves[ind - 1] for ind in allNodes]
+ host.setSlaves(newslaves)
+ self.noProcesses = len(newslaves)
- Section.__init__(self, parset, \
- 'Appl/CEP/CS1/CS1_InputSection', \
- host = host, \
- buildvar = 'gnu64_mpich-opt')
+ Section.__init__(self, parset, \
+ 'Appl/CEP/CS1/CS1_InputSection', \
+ host = host, \
+ buildvar = 'gnu64_mpich-opt')
- self.parset['Input.InputNodes'] = inputIndices
- self.parset['Input.OutputNodes'] = outputIndices
+ self.parset['Input.InputNodes'] = inputIndices
+ self.parset['Input.OutputNodes'] = outputIndices
- bglprocIPs = [newslaves[j].getExtIP() for j in outputIndices]
- self.parset['Connections.Input_BGLProc.ServerHosts'] = '[' + ','.join(bglprocIPs) + ']'
+ bglprocIPs = [newslaves[j].getExtIP() for j in outputIndices]
+ self.parset['Connections.Input_BGLProc.ServerHosts'] = '[' + ','.join(bglprocIPs) + ']'
def run(self, runlog, noRuns, runCmd = None):
Section.run(self, runlog, noRuns, runCmd)
@@ -113,18 +116,19 @@ class StorageSection(Section):
def __init__(self, parset, host):
nSubbands = parset.getInt32('Observation.NSubbands')
- nSubbandsPerCell = parset.getInt32('General.SubbandsPerPset') * parset.getInt32('BGLProc.PsetsPerCell')
- if not nSubbands % nSubbandsPerCell == 0:
- raise Exception('Not a integer number of subbands per compute cells!')
+ nSubbandsPerCell = parset.getInt32('General.SubbandsPerPset')
+ nPsetsPerStorage = parset.getInt32('Storage.PsetsPerStorage');
+ if not nSubbands % (nSubbandsPerCell * nPsetsPerStorage) == 0:
+ raise Exception('Not a integer number of subbands per storage node!')
- self.noProcesses = nSubbands / nSubbandsPerCell
+ self.noProcesses = nSubbands / (nSubbandsPerCell * nPsetsPerStorage)
Section.__init__(self, parset, \
'Appl/CEP/CS1/CS1_Storage', \
host = host, \
buildvar = 'gnu32_openmpi-opt')
- storageIPs = [s.getExtIP() for s in self.host.getSlaves(self.noProcesses)]
+ storageIPs = [s.getExtIP() for s in self.host.getSlaves(self.noProcesses * nPsetsPerStorage)]
self.parset['Connections.BGLProc_Storage.ServerHosts'] = '[' + ','.join(storageIPs) + ']'
@@ -135,8 +139,10 @@ class BGLProcSection(Section):
nSubbands = parset.getInt32('Observation.NSubbands')
nSubbandsPerCell = parset.getInt32('General.SubbandsPerPset') * parset.getInt32('BGLProc.PsetsPerCell')
+
if not nSubbands % nSubbandsPerCell == 0:
raise Exception('Not a integer number of compute cells!')
+
nCells = nSubbands / nSubbandsPerCell
self.noProcesses = int(nCells) * parset.getInt32('BGLProc.NodesPerPset') * parset.getInt32('BGLProc.PsetsPerCell')
self.noProcesses = 256 # The calculation above is not correct, because some ranks aren't used
--
GitLab