Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
Cluster Computing: UpdraftCharles ReidScientific Computing Summer WorkshopJune 29, 2010
Wednesday, June 30, 2010
Wednesday, June 30, 2010
Updraft Cluster: Hardware
• 256 Dual Quad-Core Nodes
• 2048 Cores
• 2.8 GHz Intel Xeon Processors
• 16 GB memory per node
• Qlogic Infiniband network interconnect
• Ethernet network interconnect
• 256 Dual Quad-Core Nodes
• 2048 Cores
• 2.8 GHz Intel Xeon Processors
• 16 GB memory per node
• Qlogic Infiniband network interconnect
• Ethernet network interconnect
Wednesday, June 30, 2010
Updraft Cluster: Hardware
• 256 Dual Quad-Core Nodes
• 2048 Cores
• 2.8 GHz Intel Xeon Processors
• 16 GB memory per node
• Qlogic Infiniband network interconnect
• Ethernet network interconnect
• 256 Dual Quad-Core Nodes
• 2048 Cores
• 2.8 GHz Intel Xeon Processors
• 16 GB memory per node
• Qlogic Infiniband network interconnect
• Ethernet network interconnect
Wednesday, June 30, 2010
Node Networking
• Networking “flavors” for nodes:
• Ethernet
• Infiniband
Wednesday, June 30, 2010
Node Networking
• Networking “flavors” for nodes:
• Ethernet
• Infiniband
Wednesday, June 30, 2010
Head Nodes
• Updraft 1 (updraft1.chpc.utah.edu)
• Updraft 2 (updraft2.chpc.utah.edu)
• NOT FOR COMPUTATION!!!
• Illustrative Example
Wednesday, June 30, 2010
Updraft Cluster: Hardware
• 256 Dual Quad-Core Nodes
• 2048 Cores
• 2.8 GHz Intel Xeon Processors
• 16 GB memory per node
• Qlogic Infiniband interconnect
• Gigabit ethernet interconnect
Wednesday, June 30, 2010
Simple Parallel Example
• What actually happens when you run programs on multiple cores?
Wednesday, June 30, 2010
Golden Rule:
• The one with the gold makes the rules
Wednesday, June 30, 2010
Wednesday, June 30, 2010
Funding for Updraft:
CSAFEICSE
Others
Wednesday, June 30, 2010
Updraft Policies
• ICSE and CSAFE users: running big jobs
• Updraft designed for big jobs
• High throughput vs. High capacity
• Preemption: some jobs can preempt other jobs
• QOS: your “quality of service” determines whether you can pre-empt other jobs
Wednesday, June 30, 2010
Updraft Policies: Allocations
• Allocation = how much time you get on Updraft
• Every CPU hour run decreases your allocation
• Allocations given for each quarterJan-Mar, Apr-Jun, Jul-Sep, Sep-Dec
• Allocations distributed/grouped by P.I. (primary investigators), not on individual basis
• CHPC website: current allocation statushttp://www.chpc.utah.edu
Wednesday, June 30, 2010
Updraft Policies: Preemptors
QOS Priority
bigrun+++
(preemptor)NOT FOR SMALL JOBS!!!
Jobs > 512 nodes
uintah ++(preemptor)
Must be in Uintah group
general++
(preemptor) Anyone
Wednesday, June 30, 2010
Updraft Policies: Preemptees
QOS Priority
preemptable+
(preemptable)Charges allocation
at 1/4 the rate
freecycle0
(preemptable)Doesn’t use/require
any allocation
Wednesday, June 30, 2010
Updraft Policies:DATs
• Dedicated Access Time
• 48 hours to access 100% of Updraft
• 1st week of the month - Uintah DAT
• 3rd week of the month - Uintah DAT
• 4th week of the month - General DAT
• How to find out when DATs are?
Wednesday, June 30, 2010
Wednesday, June 30, 2010
Updraft Queue System
• Queue system: schedules and manages jobs and resources
• Other systems: Moab, Maui, etc.
• PBS scripts - see CHPC website for examples
• Submitting jobs - qsub
• Deleting jobs - qdel
• Checking on jobs - checkjob
• Showing entire queue - showq / qstat
Wednesday, June 30, 2010
• When you run a job, where do you put stuff?
• NFS (Network file system)
• Uintah disk space
• Scratch & temp disk space
Updraft Hard Disks
Wednesday, June 30, 2010
• Slowest... 100 - 200 MB/s
• NOT for dumping simulation output!!!
• Home directories:/uufs/chpc.utah.edu/common/home
• CRSim group space:/uufs/chpc.utah.edu/common/home/crsim_grp
• ICSE group space:/uufs/chpc.utah.edu/common/home/icse_grp
Updraft Hard Disks: NFS
Wednesday, June 30, 2010
• Faster access... for executables and code
• NOT for simulation output!!!
• Limited disk space for code development/execution
• Only accessible from updraft2.chpc.utah.edu head node
• Location:/uufs/updraft.arches/common/uintah/homebrew
Updraft Hard Disks: Uintah Disk Space
Wednesday, June 30, 2010
• This IS for simulation output!!!
• Disk speeds are faster: 200-800 MB/s
• WARNING: Cleaned monthly/at random!
• Uintah scratch space:/scratch/uintah
• General scratch space:/scratch/general
• Node scratch space (very small):/tmp
Updraft Hard Disks: Scratch/Temp Disk Space
Wednesday, June 30, 2010
• Disk usage summaries are emailed weekly
• Email [email protected] to be added to email list for disk usage summaries
• disk usage (for files/folders): du
• disk free space (for file systems): df
Updraft Hard Disks: Getting Disk Info
Wednesday, June 30, 2010
• Compilers
• GCC compiler suite (gcc)
• Intel compiler suite (icc)
• Pathscale compilers (pathcc)
• Portland group compilers (pgcc)
Updraft Software
Wednesday, June 30, 2010
• MPI
• MPICH
• MVAPICH/MVAPICH2
• OpenMPI
• Qlogic
Updraft Software
Wednesday, June 30, 2010
• Linear algebra packages
• ATLAS
• BLAS
• LAPACK/ScaLAPACK
• IMKL (Intel Math Kernel Library)
• Hypre
• PETSc
Updraft Software
Wednesday, June 30, 2010
• Full list of software: http://www.chpc.utah.edu
• Presentations (many cover software mentioned here): http://www.chpc.utah.edu
• Locations for stuff:/uufs/chpc.utah.edu/sys/pkg/uufs/updraft.arches/sys/pkg
• Alternatively: build your own stuff $HOME/pkg
Updraft Software
Subject of next workshop!
Wednesday, June 30, 2010
Wednesday, June 30, 2010
$ system_profiler
$ cat /proc/cpuinfo
$ top
$ who
$ finger
$ showres
$ qsub
$ qdel
$ checkjob
$ qstat
$ showq
$ du -h
$ df -h
Wednesday, June 30, 2010
★ http://www.chpc.utah.edu
★Cluster Hardware:http://chpc.utah.edu/docs/manuals/hardware/
★Updraft User’s Guide:http://chpc.utah.edu/docs/manuals/user_guides/updraft
★Queue system:http://chpc.utah.edu/docs/manuals/user_guides/updraft/#batch
★Available Software:http://chpc.utah.edu/docs/manuals/software/
★Presentations:http://chpc.utah.edu/docs/presentations/
★CHPC Wiki:https://wiki.chpc.utah.edu/
Wednesday, June 30, 2010