[Swift-commit] r2797 - in usertools/cio: . libexec libexec/falkon
noreply at svn.ci.uchicago.edu
noreply at svn.ci.uchicago.edu
Wed Apr 1 16:26:00 CDT 2009
Author: aespinosa
Date: 2009-04-01 16:26:00 -0500 (Wed, 01 Apr 2009)
New Revision: 2797
Added:
usertools/cio/libexec/
usertools/cio/libexec/falkon/
usertools/cio/libexec/falkon/runworker-bgp.sh
usertools/cio/libexec/hashserver.rb
Removed:
usertools/cio/libexec/falkon/ciologic-bgp.sh
Modified:
usertools/cio/libexec/falkon/cobaltrun.job
usertools/cio/libexec/falkon/falkon-start-bgp-cio.sh
Log:
Inserted libexec directories. Scripts that are
indirectly called are placed in here
Copied: usertools/cio/libexec/falkon (from rev 2793, usertools/cio/falkon)
Deleted: usertools/cio/libexec/falkon/ciologic-bgp.sh
===================================================================
--- usertools/cio/falkon/ciologic-bgp.sh 2009-04-01 01:26:56 UTC (rev 2793)
+++ usertools/cio/libexec/falkon/ciologic-bgp.sh 2009-04-01 21:26:00 UTC (rev 2797)
@@ -1,136 +0,0 @@
-#!/fuse/bin/bash
-
-CIOROOT=/home/espinosa/cio
-ERROR_LOG=/dev/null
-
-if [ $# -lt 7 ]; then
- cat << EOF
-Usage: $0 <FALKON_IP> <TCPCORE_PORT1> <TCPCORE_PORT2> <NUM_WORKERS> <USER_ID> <FALKON_JOB_ID> <FALKON_ROOT_PATH> \
- <PART_SIZE> <DATA_NUM> <STRIPE_SIZE>
-Example: $0 192.168.1.254 55000 55001 4 iraicu 0001 [FALKON_ROOT] \
- 1024 1 1"
-EOF
- exit 2
-fi
-
-# Initiate the IP stack of the Torus network
-# -aespinosa
-
-/home/iskra/ZeptoOS/packages/cnip/prebuilt/cn-ipfwd.2409 &
-
-while [ ! -f /tmp/ifconfig.cmd ]; do
- sleep 1
-done
-
-. /tmp/ifconfig.cmd
-
-RANK=`/home/espinosa/bin/rank`
-
-# size of the allocated partition. also known as the total number of CNs
-PART_SIZE=$8
-# number of data CNs in partition
-DATA_NUM=$9
-STR=${10}
-
-ROLE=`/home/espinosa/bin/isdata $PART_SIZE $DATA_NUM $STR`
-ISDATA=$ROLE
-tmp=`/home/espinosa/bin/getdata $PART_SIZE $DATA_NUM $STR`
-DATACN=`/fuse/gpfs/home/espinosa/bin/ipcal 10.128.0.0 $tmp`
-
-if [ $RANK -eq 0 ]; then
- # Run Central Hash Service
- $CIOROOT/tools/hashserver.rb
-elif [ "$ISDATA" != 'app' ]; then
- if [ "$ROLE" = 'head' ]; then
- # Run head IFS service
- $CIOROOT/tools/ifshead.sh $DATACN $STR >&2
- else
- # Contribute stripes to the IFS service
- $CIOROOT/tools/ifsslave.sh $DATACN >&2
- fi
-else
- /bin/mkdir /dataifs
- export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/lib:/fuse/lib:/usr/lib
- sleep 30
- # Contact and mount assigned IFS head
- $CIOROOT/tools/ifsmount.sh $DATACN >&2
-
- echo "worker started..." >> ${ERROR_LOG}
-
- # Regular Falkon worker startup script
-
- FALKON_SERVICE_IP=$1
- TCPCORE_PORT1=$2
- TCPCORE_PORT2=$3
- NUM_WORKERS=$4
- USER_ID=$5
- FALKON_JOB_ID=$6
- FALKON_ROOT_PATH=$7
-
-
- FALKON_HOME=${FALKON_ROOT_PATH}/users/${USER_ID}/${FALKON_JOB_ID}
-
- echo "FALKON_HOME = ${FALKON_HOME}" >> ${ERROR_LOG}
-
- #FALKON_HOME=/home/falkon/falkon
-
- HOME=${FALKON_ROOT_PATH}
- echo "HOME = ${HOME}" >> ${ERROR_LOG}
- #HOME=${FALKON_HOME}
-
- if [ -z "${FALKON_HOME}" ]; then
- echo "ERROR: environment variable FALKON_HOME not defined"
- echo "ERROR: environment variable FALKON_HOME not defined" >> ${ERROR_LOG}
- exit 1
- fi
-
- if [ ! -d "${FALKON_HOME}" ]; then
- echo "ERROR: invalid FALKON_HOME set: $FALKON_HOME"
- echo "ERROR: invalid FALKON_HOME set: $FALKON_HOME" >> ${ERROR_LOG}
- exit 1
- fi
-
- cd ${FALKON_HOME}
- source falkon.env.bgp-cn
-
-
-
- HOME_PATH=`pwd`
- export MACH_ID=`uname -i`
- RAND_ID2=`/home/espinosa/bin/rank`
-
- a0=${RAND_ID2%%=*}; rest=${RAND_ID2#*=}
- RAND_ID=${rest%%=*};
-
- #PLATFORM=`uname -m`
-
-
-
- EXP_START=`date +%Y.%m.%d_%H.%M.%S`
-
-
- for ((b=1; b < ${NUM_WORKERS} ; b++))
- do
- #LOG_FILE=${FALKON_LOGS}/worker/${RAND_ID}.${b}.${EXP_START}.txt
- LOG_FILE=/dev/null
-
- #${FALKON_WORKER_HOME}/src-c/BGexec_${PLATFORM} ${FALKON_SERVICE_IP} ${TCPCORE_PORT1} ${TCPCORE_PORT2} -perf >> "${LOG_FILE}" 2>&1 &
- #echo "starting .... ${FALKON_WORKER_HOME}/src-c/BGexec ${FALKON_SERVICE_IP} ${TCPCORE_PORT1} ${TCPCORE_PORT2} -perf"
- echo "starting worker ${RAND_ID} CPU ${b}..." >> ${ERROR_LOG}
- ${FALKON_WORKER_HOME}/src-c/BGexec ${FALKON_SERVICE_IP} ${TCPCORE_PORT1} ${TCPCORE_PORT2} -perf >> ${LOG_FILE} 2>&1 &
- #${HOME_PATH}/src-c/BGexec_${PLATFORM} ${FALKON_SERVICE_IP} 55000 55001 -debug
- /bin/sleep 1
- done
-
- #LOG_FILE=${FALKON_LOGS}/worker/${RAND_ID}.${NUM_WORKERS}.${EXP_START}.txt
- LOG_FILE=/dev/null
- echo "starting worker ${RAND_ID} CPU ${NUM_WORKERS}..." >> ${ERROR_LOG}
-
- #${FALKON_WORKER_HOME}/src-c/BGexec_${PLATFORM} ${FALKON_SERVICE_IP} ${TCPCORE_PORT1} ${TCPCORE_PORT2} -perf >> "${LOG_FILE}" 2>&1
- #echo "starting .... ${FALKON_WORKER_HOME}/src-c/BGexec ${FALKON_SERVICE_IP} ${TCPCORE_PORT1} ${TCPCORE_PORT2} -perf"
-
- ${FALKON_WORKER_HOME}/src-c/BGexec ${FALKON_SERVICE_IP} ${TCPCORE_PORT1} ${TCPCORE_PORT2} -perf >> ${LOG_FILE} 2>&1
-
-
- echo "finished!" >> ${ERROR_LOG}
-fi
Modified: usertools/cio/libexec/falkon/cobaltrun.job
===================================================================
--- usertools/cio/falkon/cobaltrun.job 2009-04-01 01:26:56 UTC (rev 2793)
+++ usertools/cio/libexec/falkon/cobaltrun.job 2009-04-01 21:26:00 UTC (rev 2797)
@@ -1 +1 @@
-./falkon-start-bgp-logging.sh default 64 60 4
+./falkon-start-bgp-cio.sh default 64 15 4 1 3
Modified: usertools/cio/libexec/falkon/falkon-start-bgp-cio.sh
===================================================================
--- usertools/cio/falkon/falkon-start-bgp-cio.sh 2009-04-01 01:26:56 UTC (rev 2793)
+++ usertools/cio/libexec/falkon/falkon-start-bgp-cio.sh 2009-04-01 21:26:00 UTC (rev 2797)
@@ -1,7 +1,5 @@
#!/bin/bash
-CIOROOT=/home/espinosa/cio
-
if [ $# -lt 3 ]; then
cat << EOF
Usage: $0 <QueueName> <NumNodes> <WallTime>
@@ -17,6 +15,14 @@
echo "ERROR: environment variable FALKON_HOME not defined"
exit 1
fi
+if [ -z $CIOROOT ]; then
+ echo "ERROR: CIOROOT env not defined"
+ exit 1
+fi
+if [[ $CIOARCH != "bgp" ]]; then
+ echo "ERROR: Wrong architecture. Must be bgp"
+ exit 1
+fi
PROFILE_NAME="zeptocn"
QUEUE_NAME=$1
@@ -30,17 +36,17 @@
WORKERS_PER_NODE=4
DATA_NUM=1
STR=1
-if [ -n $4 ];then
+if [ -n "$4" ];then
WORKERS_PER_NODE=$4
fi
-if [ -n $5 ];then
+if [ -n "$5" ];then
DATA_NUM=$5
fi
-if [ -n $6 ];then
+if [ -n "$6" ];then
STR=$6
+ echo true
fi
-
cp ${FALKON_HOME}/bin/zoid-user-script.sh ${HOME}/zoid-user-script.sh
chmod +x ${HOME}/zoid-user-script.sh
@@ -55,13 +61,14 @@
-WORKER_SCRIPT=$CIOROOT/falkon/ciologic-bgp.sh
+WORKER_SCRIPT=$CIOROOT/bin/ciologic-$CIOARCH.sh
ZOID_JOB_ID=`cqsub -q ${QUEUE_NAME} -k ${PROFILE_NAME} -C ${HOME} -t \
${MAX_TIME_MIN} -n ${NUM_NODES} \
- -e LD_LIBRARY_PATH=/lib:/fuse/lib:/fuse/usr/lib \
+ -e LD_LIBRARY_PATH=/lib:/fuse/lib:/fuse/usr/lib:CIOROOT=$CIOROOT:CIOARCH=$CIOARCH:CHIRPROOT=$CHIRPROOT \
/bgsys/linux/1.2.020080512/bin/bash $WORKER_SCRIPT \
${SERVICE_IP} ${SERVICE_PORT1} ${SERVICE_PORT2} ${WORKERS_PER_NODE} \
- ${USER} ${FALKON_JOB_ID} ${FALKON_ROOT} ${NUM_NODES} ${DATA_NUM} ${STR}`
+ ${USER} ${FALKON_JOB_ID} ${FALKON_ROOT} \
+ $NUM_NODES $DATA_NUM $STR` # MTIO parameters
EXIT_CODE=$?
Added: usertools/cio/libexec/falkon/runworker-bgp.sh
===================================================================
--- usertools/cio/libexec/falkon/runworker-bgp.sh (rev 0)
+++ usertools/cio/libexec/falkon/runworker-bgp.sh 2009-04-01 21:26:00 UTC (rev 2797)
@@ -0,0 +1,100 @@
+#!/fuse/bin/bash
+
+#source /home/iraicu/.bashrc
+
+#ERROR_LOG=${FALKON_ROOT_PATH}/users/error.log
+#ERROR_LOG=/home/falkon/users/error-workers.log
+ERROR_LOG=/dev/null
+ echo "worker started..." >> ${ERROR_LOG}
+
+
+ if [ -z "$7" ]; then
+ echo "usage: $0 <FALKON_IP> <TCPCORE_PORT1> <TCPCORE_PORT2> <NUM_WORKERS> <USER_ID> <FALKON_JOB_ID> <FALKON_ROOT_PATH>"
+ echo "usage: $0 <FALKON_IP> <TCPCORE_PORT1> <TCPCORE_PORT2> <NUM_WORKERS> <USER_ID> <FALKON_JOB_ID> <FALKON_ROOT_PATH>" >> ${ERROR_LOG}
+ echo "$0 192.168.1.254 55000 55001 4 iraicu 0001"
+ echo "$0 192.168.1.254 55000 55001 4 iraicu 0001" >> ${ERROR_LOG}
+ exit 2
+ fi
+
+FALKON_SERVICE_IP=$1
+TCPCORE_PORT1=$2
+TCPCORE_PORT2=$3
+NUM_WORKERS=$4
+USER_ID=$5
+FALKON_JOB_ID=$6
+FALKON_ROOT_PATH=$7
+
+
+ ln -s /fuse/gpfs1 /gpfs1
+ #FALKON_HOME=/gpfs1/falkon/falkon
+ #HOME=/gpfs1/falkon
+
+
+
+ #echo "Worker for job id ${ZOID_JOB_ID} starting..." >> ${ERROR_LOG}
+
+ FALKON_HOME=${FALKON_ROOT_PATH}/users/${USER_ID}/${FALKON_JOB_ID}
+
+ echo "FALKON_HOME = ${FALKON_HOME}" >> ${ERROR_LOG}
+
+ #FALKON_HOME=/home/falkon/falkon
+
+ HOME=${FALKON_ROOT_PATH}
+ echo "HOME = ${HOME}" >> ${ERROR_LOG}
+ #HOME=${FALKON_HOME}
+
+ if [ -z "${FALKON_HOME}" ]; then
+ echo "ERROR: environment variable FALKON_HOME not defined"
+ echo "ERROR: environment variable FALKON_HOME not defined" >> ${ERROR_LOG}
+ exit 1
+ fi
+
+ if [ ! -d "${FALKON_HOME}" ]; then
+ echo "ERROR: invalid FALKON_HOME set: $FALKON_HOME"
+ echo "ERROR: invalid FALKON_HOME set: $FALKON_HOME" >> ${ERROR_LOG}
+ exit 1
+ fi
+
+cd ${FALKON_HOME}
+source falkon.env.bgp-cn
+
+
+
+
+HOME_PATH=`pwd`
+#export MACH_ID=`uname -i`
+RAND_ID2=`grep BG_RANK /proc/personality.sh`
+
+a0=${RAND_ID2%%=*}; rest=${RAND_ID2#*=}
+RAND_ID=${rest%%=*};
+
+#PLATFORM=`uname -m`
+
+
+
+EXP_START=`date +%Y.%m.%d_%H.%M.%S`
+
+for ((b=1; b < ${NUM_WORKERS} ; b++))
+do
+ #LOG_FILE=${FALKON_LOGS}/worker/${RAND_ID}.${b}.${EXP_START}.txt
+ LOG_FILE=/dev/null
+
+ #${FALKON_WORKER_HOME}/src-c/BGexec_${PLATFORM} ${FALKON_SERVICE_IP} ${TCPCORE_PORT1} ${TCPCORE_PORT2} -perf >> "${LOG_FILE}" 2>&1 &
+ #echo "starting .... ${FALKON_WORKER_HOME}/src-c/BGexec ${FALKON_SERVICE_IP} ${TCPCORE_PORT1} ${TCPCORE_PORT2} -perf"
+ echo "starting worker ${RAND_ID} CPU ${b}..." >> ${ERROR_LOG}
+ /home/zzhang/falkon/worker/src-c/BGexec_new ${FALKON_SERVICE_IP} ${TCPCORE_PORT1} ${TCPCORE_PORT2} -perf >> ${LOG_FILE} 2>&1 &
+ #${HOME_PATH}/src-c/BGexec_${PLATFORM} ${FALKON_SERVICE_IP} 55000 55001 -debug
+/bin/sleep 1
+done
+
+ #LOG_FILE=${FALKON_LOGS}/worker/${RAND_ID}.${NUM_WORKERS}.${EXP_START}.txt
+ LOG_FILE=/dev/null
+ echo "starting worker ${RAND_ID} CPU ${NUM_WORKERS}..." >> ${ERROR_LOG}
+
+ #${FALKON_WORKER_HOME}/src-c/BGexec_${PLATFORM} ${FALKON_SERVICE_IP} ${TCPCORE_PORT1} ${TCPCORE_PORT2} -perf >> "${LOG_FILE}" 2>&1
+ #echo "starting .... ${FALKON_WORKER_HOME}/src-c/BGexec ${FALKON_SERVICE_IP} ${TCPCORE_PORT1} ${TCPCORE_PORT2} -perf"
+ /home/zzhang/falkon/worker/src-c/BGexec_new ${FALKON_SERVICE_IP} ${TCPCORE_PORT1} ${TCPCORE_PORT2} -perf >> ${LOG_FILE} 2>&1
+
+
+ echo "finished!" >> ${ERROR_LOG}
+
Property changes on: usertools/cio/libexec/falkon/runworker-bgp.sh
___________________________________________________________________
Name: svn:executable
+ *
Copied: usertools/cio/libexec/hashserver.rb (from rev 2793, usertools/cio/tools/hashserver.rb)
===================================================================
--- usertools/cio/libexec/hashserver.rb (rev 0)
+++ usertools/cio/libexec/hashserver.rb 2009-04-01 21:26:00 UTC (rev 2797)
@@ -0,0 +1,26 @@
+#!/home/espinosa/local/bin/ruby
+
+require 'drb'
+
+class HashServer
+ attr_reader :file
+
+ def initialize
+ @file = Hash.new { |hash, key| hash[key] = Array.new }
+ end
+
+ def put(fname, rank)
+ @file[fname] << rank
+ end
+
+ def get(fname)
+ @file[fname][ rand( @file[fname].size ) ]
+ end
+
+end
+
+server = HashServer.new
+
+DRb.start_service('druby://*:9000', server)
+DRb.thread.join
+
More information about the Swift-commit
mailing list