DefaultSerialQueue "msg-sgehead1 cpu-serial" DefaultClusterQueue "msg-sgehead1 cpu-cluster" DefaultThreadedQueue "msg-sgehead1 cpu-threaded" DefaultGPUQueue "msg-sgehead1 gpu-threaded" # If running directly on gpu, make the "gpu1 8800 GT" queue the default # destination for GPU jobs. IFHOST { "gpu1.ucsf.edu" } DefaultGPUQueue "gpu1 8800 GT" ENDIF # Collect all the MSG multiprocessor hosts. IFHOST { "lbox1.ucsf.edu" "lbox2.ucsf.edu" "venus.ucsf.edu" } HOST ProcessControl "simple" DefaultProcessorCount 8 StatusCommand "w" ENDIF IFHOST { "blue.ucsf.edu" "ganymede.ucsf.edu" "neptune.ucsf.edu" "pluto.ucsf.edu" } HOST ProcessControl "simple" DefaultProcessorCount 2 StatusCommand "w" ENDIF # Define a generic host. HOST ProcessControl "simple" DefaultProcessorCount 1 StatusCommand "w" # Collect all the MSG compute queues. QUEUE Name "msg-sgehead1 cpu-serial" Description "Submits a serial job to one node of agard.q which has 10 16-core 2.6 GHz Xeon E5-2670 systems each with 132 GB of RAM." AcceptsSerial true AcceptsCluster false AcceptsThreaded false AcceptsGPU false # SGE will set NSLOTS so don't pass a processor count to the applications. SetProcessorCountForApp false ProcessControl "simple" DefaultProcessorCount 1 DefaultMPIRunCommand "/software/openmpi-1.6/bin/mpiexec" SubmissionCommand "ssh -nx sgehead.msg.ucsf.edu /opt/sge/bin/lx24-amd64/qsub -pe cpu 1 -l qname=agard.q -l mem_free=8G -S /bin/sh" StatusCommand "ssh -nx sgehead.msg.ucsf.edu /opt/sge/bin/lx24-amd64/qstat" # Skip the unneeded call to ssh on hosts that are submit hosts. IFHOST { "sgehead.msg.ucsf.edu" "node-9-01.c.msg.ucsf.edu" "node-9-02.c.msg.ucsf.edu" "node-9-03.c.msg.ucsf.edu" "node-9-04.c.msg.ucsf.edu" "node-9-05.c.msg.ucsf.edu" "node-9-06.c.msg.ucsf.edu" "node-9-07.c.msg.ucsf.edu" "node-9-08.c.msg.ucsf.edu" "node-9-09.c.msg.ucsf.edu" "node-9-10.c.msg.ucsf.edu" "nvgpu-1-1.c.msg.ucsf.edu" "nvgpu-1-2.c.msg.ucsf.edu" "nvgpu-1-3.c.msg.ucsf.edu" "nvgpu-1-4.c.msg.ucsf.edu" "nvgpu-1-5.c.msg.ucsf.edu" "nvgpu-1-6.c.msg.ucsf.edu" "nvgpu-1-7.c.msg.ucsf.edu" "nvgpu-1-8.c.msg.ucsf.edu" "nvgpu-2-1.c.msg.ucsf.edu" "nvgpu-2-2.c.msg.ucsf.edu" "nvgpu-2-3.c.msg.ucsf.edu" "nvgpu-2-4.c.msg.ucsf.edu" "nvgpu-2-5.c.msg.ucsf.edu" "nvgpu-2-6.c.msg.ucsf.edu" "nvgpu-2-7.c.msg.ucsf.edu" "nvgpu-2-8.c.msg.ucsf.edu" } SubmissionCommand "/opt/sge/bin/lx24-amd64/qsub -pe cpu 1 -l qname=agard.q -l mem_free=8G -S /bin/sh" StatusCommand "/opt/sge/bin/lx24-amd64/qstat" ENDIF QUEUE Name "msg-sgehead1 cpu-threaded" Description "Submits a parallel job to one node of agard.q which has 10 16-core 2.6 GHz Xeon E5-2670 systems each with 132 GB of RAM." AcceptsSerial false AcceptsCluster true AcceptsThreaded true AcceptsGPU false # SGE will set NSLOTS so don't pass a processor count to the applications. SetProcessorCountForApp false ProcessControl "simple" DefaultProcessorCount 8 DefaultMPIRunCommand "/software/openmpi-1.6/bin/mpiexec" SubmissionCommand "ssh -nx sgehead.msg.ucsf.edu /opt/sge/bin/lx24-amd64/qsub -pe cpu 1-%p -l qname=agard.q -l mem_free=50G -S /bin/sh" StatusCommand "ssh -nx sgehead.msg.ucsf.edu /opt/sge/bin/lx24-amd64/qstat" # Skip the unneeded call to ssh on hosts that are submit hosts. IFHOST { "sgehead.msg.ucsf.edu" "node-9-01.c.msg.ucsf.edu" "node-9-02.c.msg.ucsf.edu" "node-9-03.c.msg.ucsf.edu" "node-9-04.c.msg.ucsf.edu" "node-9-05.c.msg.ucsf.edu" "node-9-06.c.msg.ucsf.edu" "node-9-07.c.msg.ucsf.edu" "node-9-08.c.msg.ucsf.edu" "node-9-09.c.msg.ucsf.edu" "node-9-10.c.msg.ucsf.edu" "nvgpu-1-1.c.msg.ucsf.edu" "nvgpu-1-2.c.msg.ucsf.edu" "nvgpu-1-3.c.msg.ucsf.edu" "nvgpu-1-4.c.msg.ucsf.edu" "nvgpu-1-5.c.msg.ucsf.edu" "nvgpu-1-6.c.msg.ucsf.edu" "nvgpu-1-7.c.msg.ucsf.edu" "nvgpu-1-8.c.msg.ucsf.edu" "nvgpu-2-1.c.msg.ucsf.edu" "nvgpu-2-2.c.msg.ucsf.edu" "nvgpu-2-3.c.msg.ucsf.edu" "nvgpu-2-4.c.msg.ucsf.edu" "nvgpu-2-5.c.msg.ucsf.edu" "nvgpu-2-6.c.msg.ucsf.edu" "nvgpu-2-7.c.msg.ucsf.edu" "nvgpu-2-8.c.msg.ucsf.edu" } SubmissionCommand "/opt/sge/bin/lx24-amd64/qsub -pe cpu 1-%p -l qname=agard.q -l mem_free=50G -S /bin/sh" StatusCommand "/opt/sge/bin/lx24-amd64/qstat" ENDIF QUEUE Name "msg-sgehead1 cpu-cluster" Description "Submits a parallel job to one or more nodes of agard.q which has 10 16-core 2.6 GHz Xeon E5-2670 systems each with 132 GB of RAM." AcceptsSerial false AcceptsCluster true AcceptsThreaded false AcceptsGPU false # SGE will set NSLOTS so don't pass a processor count to the applications. SetProcessorCountForApp false ProcessControl "openmpi" DefaultProcessorCount 20 DefaultMPIRunCommand "/software/openmpi-1.6/bin/mpiexec" SubmissionCommand "ssh -nx sgehead.msg.ucsf.edu /opt/sge/bin/lx24-amd64/qsub -pe openmpi 1-%p -l qname=agard.q -l mem_free=80G -S /bin/sh" StatusCommand "ssh -nx sgehead.msg.ucsf.edu /opt/sge/bin/lx24-amd64/qstat" # Skip the unneeded call to ssh on hosts that are submit hosts. IFHOST { "sgehead.msg.ucsf.edu" "node-9-01.c.msg.ucsf.edu" "node-9-02.c.msg.ucsf.edu" "node-9-03.c.msg.ucsf.edu" "node-9-04.c.msg.ucsf.edu" "node-9-05.c.msg.ucsf.edu" "node-9-06.c.msg.ucsf.edu" "node-9-07.c.msg.ucsf.edu" "node-9-08.c.msg.ucsf.edu" "node-9-09.c.msg.ucsf.edu" "node-9-10.c.msg.ucsf.edu" "nvgpu-1-1.c.msg.ucsf.edu" "nvgpu-1-2.c.msg.ucsf.edu" "nvgpu-1-3.c.msg.ucsf.edu" "nvgpu-1-4.c.msg.ucsf.edu" "nvgpu-1-5.c.msg.ucsf.edu" "nvgpu-1-6.c.msg.ucsf.edu" "nvgpu-1-7.c.msg.ucsf.edu" "nvgpu-1-8.c.msg.ucsf.edu" "nvgpu-2-1.c.msg.ucsf.edu" "nvgpu-2-2.c.msg.ucsf.edu" "nvgpu-2-3.c.msg.ucsf.edu" "nvgpu-2-4.c.msg.ucsf.edu" "nvgpu-2-5.c.msg.ucsf.edu" "nvgpu-2-6.c.msg.ucsf.edu" "nvgpu-2-7.c.msg.ucsf.edu" "nvgpu-2-8.c.msg.ucsf.edu" } SubmissionCommand "/opt/sge/bin/lx24-amd64/qsub -pe openmpi 1-%p -l qname=agard.q -l mem_free=80G -S /bin/sh" StatusCommand "/opt/sge/bin/lx24-amd64/qstat" ENDIF QUEUE Name "msg-sgehead1 gpu-threaded" Description "Submits a GPU job to one node of nvgpu.q which has 16 12-core 2.67 GHz Xeon X5650 systems each with 46 GB of RAM and 8 GTX 580." AcceptsSerial false AcceptsCluster false AcceptsThreaded false AcceptsGPU true # SGE will set NSLOTS so don't pass a processor count to the applications. SetProcessorCountForApp false ProcessControl "simple" # The GPU reconstruction software will use all the GPUs (unless some are marked # as busy) so allocate all of the CPUs so that the queueing system thinks the # system is full. DefaultProcessorCount 12 DefaultMPIRunCommand "/software/openmpi-1.6/bin/mpiexec" SubmissionCommand "ssh -nx sgehead.msg.ucsf.edu /opt/sge/bin/lx24-amd64/qsub -pe cpu %p -l qname=nvgpu.q -l mem_free=40G -S /bin/sh" StatusCommand "ssh -nx sgehead.msg.ucsf.edu /opt/sge/bin/lx24-amd64/qstat" # Skip the unneeded call to ssh on hosts that are submit hosts. IFHOST { "sgehead.msg.ucsf.edu" "node-9-01.c.msg.ucsf.edu" "node-9-02.c.msg.ucsf.edu" "node-9-03.c.msg.ucsf.edu" "node-9-04.c.msg.ucsf.edu" "node-9-05.c.msg.ucsf.edu" "node-9-06.c.msg.ucsf.edu" "node-9-07.c.msg.ucsf.edu" "node-9-08.c.msg.ucsf.edu" "node-9-09.c.msg.ucsf.edu" "node-9-10.c.msg.ucsf.edu" "nvgpu-1-1.c.msg.ucsf.edu" "nvgpu-1-2.c.msg.ucsf.edu" "nvgpu-1-3.c.msg.ucsf.edu" "nvgpu-1-4.c.msg.ucsf.edu" "nvgpu-1-5.c.msg.ucsf.edu" "nvgpu-1-6.c.msg.ucsf.edu" "nvgpu-1-7.c.msg.ucsf.edu" "nvgpu-1-8.c.msg.ucsf.edu" "nvgpu-2-1.c.msg.ucsf.edu" "nvgpu-2-2.c.msg.ucsf.edu" "nvgpu-2-3.c.msg.ucsf.edu" "nvgpu-2-4.c.msg.ucsf.edu" "nvgpu-2-5.c.msg.ucsf.edu" "nvgpu-2-6.c.msg.ucsf.edu" "nvgpu-2-7.c.msg.ucsf.edu" "nvgpu-2-8.c.msg.ucsf.edu" } SubmissionCommand "/opt/sge/bin/lx24-amd64/qsub -pe cpu %p -l qname=nvgpu.q -l mem_free=40G -S /bin/sh" StatusCommand "/opt/sge/bin/lx24-amd64/qstat" ENDIF QUEUE Name "msg-sgehead1 gpu-cluster" Description "Submits a GPU job to one or more nodes of nvgpu.q which has 16 12-core 2.67 GHz Xeon X5650 systems each with 46 GB of RAM and 8 GTX 580." AcceptsSerial false AcceptsCluster false AcceptsThreaded false AcceptsGPU true # SGE will set NSLOTS so don't pass a processor count to the applications. SetProcessorCountForApp false ProcessControl "openmpi" # The GPU reconstruction software will use all the GPUs (unless some are marked # as busy) so best to use something that is a multiple of the total number of # processors on each system. DefaultProcessorCount 24 DefaultMPIRunCommand "/software/openmpi-1.6/bin/mpiexec" SubmissionCommand "ssh -nx sgehead.msg.ucsf.edu /opt/sge/bin/lx24-amd64/qsub -pe openmpi %p -l qname=nvgpu.q -l mem_free=40G -S /bin/sh" StatusCommand "ssh -nx sgehead.msg.ucsf.edu /opt/sge/bin/lx24-amd64/qstat" # Skip the unneeded call to ssh on hosts that are submit hosts. IFHOST { "sgehead.msg.ucsf.edu" "node-9-01.c.msg.ucsf.edu" "node-9-02.c.msg.ucsf.edu" "node-9-03.c.msg.ucsf.edu" "node-9-04.c.msg.ucsf.edu" "node-9-05.c.msg.ucsf.edu" "node-9-06.c.msg.ucsf.edu" "node-9-07.c.msg.ucsf.edu" "node-9-08.c.msg.ucsf.edu" "node-9-09.c.msg.ucsf.edu" "node-9-10.c.msg.ucsf.edu" "nvgpu-1-1.c.msg.ucsf.edu" "nvgpu-1-2.c.msg.ucsf.edu" "nvgpu-1-3.c.msg.ucsf.edu" "nvgpu-1-4.c.msg.ucsf.edu" "nvgpu-1-5.c.msg.ucsf.edu" "nvgpu-1-6.c.msg.ucsf.edu" "nvgpu-1-7.c.msg.ucsf.edu" "nvgpu-1-8.c.msg.ucsf.edu" "nvgpu-2-1.c.msg.ucsf.edu" "nvgpu-2-2.c.msg.ucsf.edu" "nvgpu-2-3.c.msg.ucsf.edu" "nvgpu-2-4.c.msg.ucsf.edu" "nvgpu-2-5.c.msg.ucsf.edu" "nvgpu-2-6.c.msg.ucsf.edu" "nvgpu-2-7.c.msg.ucsf.edu" "nvgpu-2-8.c.msg.ucsf.edu" } SubmissionCommand "/opt/sge/bin/lx24-amd64/qsub -pe openmpi %p -l qname=nvgpu.q -l mem_free=40G -S /bin/sh" StatusCommand "/opt/sge/bin/lx24-amd64/qstat" ENDIF QUEUE Name "gpu1 GTX 580" Description "Submits a job to 2 systems. Both have an Intel Core2 6600 (2.4 GHz), 2 GB RAM, and a Nvidia GTX 580 (1536 MB). Please use this queue only for jobs that can be accelerated by the GPUs." AcceptsSerial false AcceptsCluster false AcceptsThreaded false AcceptsGPU true ProcessControl "simple" GrecsrvRunning true GrecsrvHosts { "c0-0" } SubmissionCommand "ssh -nx gpu1.ucsf.edu /opt/gridengine/bin/lx26-amd64/qsub -S /bin/sh -q nv580.q" StatusCommand "ssh -x gpu1.ucsf.edu /opt/gridengine/bin/lx26-amd64/qstat -f" # Skip the unneeded call to ssh. IFHOST { "gpu1.ucsf.edu" } SubmissionCommand "/opt/gridengine/bin/lx26-amd64/qsub -S /bin/sh -q nv580.q" StatusCommand "/opt/gridengine/bin/lx26-amd64/qstat -f" ENDIF QUEUE Name "gpu1 8800 GT" Description "Submits a job to 2 systems. Both have an Intel P4 (3.2 GHz), 2 GB RAM, and a Nvidia 8800 GT (512 MB). Please use this queue only for jobs that can be accelerated by the GPUs." AcceptsSerial false AcceptsCluster false AcceptsThreaded false AcceptsGPU true ProcessControl "simple" GrecsrvRunning true GrecsrvHosts { "c0-4" "c0-5" } SubmissionCommand "ssh -nx gpu1.ucsf.edu /opt/gridengine/bin/lx26-amd64/qsub -S /bin/sh -q gt.q" StatusCommand "ssh -x gpu1.ucsf.edu /opt/gridengine/bin/lx26-amd64/qstat -f" # Skip the unneeded call to ssh. IFHOST { "gpu1.ucsf.edu" } SubmissionCommand "/opt/gridengine/bin/lx26-amd64/qsub -S /bin/sh -q gt.q" StatusCommand "/opt/gridengine/bin/lx26-amd64/qstat -f" ENDIF QUEUE Name "gpu1 GTX 480" Description "Submits a job to one system with an Intel Core2 6600 (2.4 GHz), 2 GB RAM, and a Nvidia GTX 480 (1536 MB). Please use this queue only for jobs that can be accelerated by the GPUs." AcceptsSerial false AcceptsCluster false AcceptsThreaded false AcceptsGPU true ProcessControl "simple" GrecsrvRunning true GrecsrvHosts { "c0-1" } SubmissionCommand "ssh -nx gpu1.ucsf.edu /opt/gridengine/bin/lx26-amd64/qsub -S /bin/sh -q nv480.q" StatusCommand "ssh -x gpu1.ucsf.edu /opt/gridengine/bin/lx26-amd64/qstat -f" # Skip the unneeded call to ssh. IFHOST { "gpu1.ucsf.edu" } SubmissionCommand "/opt/gridengine/bin/lx26-amd64/qsub -S /bin/sh -q nv480.q" StatusCommand "/opt/gridengine/bin/lx26-amd64/qstat -f" ENDIF