https://mitmath.github.io/18337/lecture3/TX-E1_Reference_Guide_020819.pdf https://supercloud.mit.edu/getting-started https://github.com/llsc-supercloud/teaching-examples
ssh [email protected] https://txe1-portal.mit.edu/
- files are not backed up
- DOWNTIME for maintenance all day on 3rd thursday of every month
LLGrid_status [opteron | xeon-e5 | gpu | all] LLfree Monitor Your Job LLstat Stop Your Job LLkill [options] <job_id_list> job_id_list Comma separated list of job number(s) Options: –version Show program’s version number and exit-h, –help Show this help message and exit -t <task_list> Specify selected job array elements to delete task_list syntax: start[-end[:step]] -u <username> Delete all jobs owned by username No effect if job_id_list is specified
use rsync or scp
scp [-rv] <src> <user>@txe1-login.mit.edu:<dest>
-r recursive -v verbose
rsync [-rlugv] <src> <user>@txe1-login. mit.edu:<dest>
-r Recursively copy files in all subdirectories -l Copy and retain symbolic links -u Skip files that are newer on the LLSC system -g Preserve group attributes (use with shared groups) -v Verbose
native slurm commands work
LLGrid wrapper script to submit a batch job or get an interactive session
Usage: LLsub command [N] [options] [– command_args] LLsub -i [full|-s slots_per_task|-N num_Nodes] [–x11] [–resv-ports count] (interactive session only) [options]
command Specify the program file to execute N Specify the requested number of tasks or processes
Options: -h Show this helper message. -V Show the version number. -c CPU_type Request the nodes with the specified CPU type. -g gpuName:count Request the specified GPU resources. count should be 2 or 4. (example: -g tesla:2) -i [full] Request an interactive session [with an exclusive node access by specifying ‘full’]. –x11 Enable X11 forwarding for interactive session. -J job_name Assigns the specified name to the job. The job name does not need to be unique. -N num_nodes Number of nodes to allocate (interactive job only) -o log_file Specify a file path. Appends the standard output of the job to the specified file. -q queue_name Specify a queue name to which a job is submitted. -Q QoS Specify a QoS for the job. -r adv_res_name Specify an advanced reservation name to which a job is submitted. (requires LLSC coordination) –resv-ports count Reserve communication ports for this job. (interactive job only) -s slots_per_task Specify a number of CPUs allocated per each task. -t task_list Specify tasks of an array job starting from a user-defined task ID. The syntax is start[-end[:step]] or a comma-separated task list [t1,t2,t3]. -T time_limit Set a limit on the total run time of the job allocation. Acceptable formats: ‘minutes’, ‘hours:minutes:seconds’, ‘days-hours’, ‘days-hours:minutes’ -w job_number | Define the job dependency list of the submitted job. job_name
Option for MPI parallel jobs: -a mpi Specify that the command is an MPI application. The mpirun command will be selected from your PATH.
If you need any other options, please send your requirement to [email protected].
LLMapReduce Usage: LLMapReduce [options]
Options: –version show program’s version number and exit -h, –help show this help message and exit –np=NPROCS Number of processes to run concurrently. Without –ndata, all data will be evenly distributed to the given number of processes. –ndata=NDATAPERTASK Number of input data to be processed per task for fine control, The default value is one. You may want to use the –np option instead if you want to distribute the entire work load to the given nProcs processes by the –np option. –distribution=DATADIST Distribution rule for the data, block or cyclic. Default is block. –mapper=MYMAPPER Specify the mapper program to execute. –input=INPATH Specify a path where the input files are or a file containing the list of the input files. –output=OUTDIRPATH Specify a directory path where the output files to be saved. –prefix=PREFIXFILENAME Specify a string to be prefixed to output file name. –subdir=USESUBDIR Specify true if data is located at sub-directories. All the data under the input directory will be scanned recursively. The same sub-directory structure will be maintained under the output directory. Default is false. –ext=OUTEXTENSION Specify a file extension for the output. Default is out. Use noext if no extension is preferred. –extSearch=SEARCHEXTENSION Specify a file extension when searching input files with the –subdir option. –delimeter=OUTDELIMETER Specify a file extension delimeter for the output. Default is the dot(.) –exclusive=EXCLUSIVEMODE Turn on the exclusive mode (true/false). The default is false. –reducer=MYREDUCER Specify the reducer program to execute. –redout=REDOUTFILENAME Output filename for the reducer [optional]. –changeDepMode=DEPENDENCYMODE Change the dependency mode. By default, the reduce job starts only when all mapper tasks are completed successfully. The alternative behavior (–changeDepMode=true) lets the reduce job start when the mapper job terminates regardless of its exit status. –keep=KEEPTEMPDIR Decide whether or not to keep the temporary .MATPRED.PID dirctory (true/false). The default is false. –apptype=APPLICATIONTYPE If your application can take multiple lines of input and output format, set apptype=mimo. By default, your application takes one line of input and output (siso). –cpuType=CPUTYPE Request compute nodes with a specific CPU type [optional]. –gpuNameCount=GPUNAMECOUNT Specify the GPU name and number of counts to be used for each task as GPU_NAME:COUNT. Currently each node has 4 tesla (K80) units. –slotsPerTask=SLOTSPERTASK Specify the number of slots(cores) per task. Default value is 1 [optional]. –slotsPerTaskType=SLOTSPERTASKTYPE Specify how the number of slots(cores) per task be applied. Default value is 1 [Map only], Other options are 2 [Both Map and Reduce] and 3 [Reduce only]. –reservation=ADV_RES_NAME Specify an advanced reservation name to which a job is submitted (requires LLSC coordination). –tempdir=TEMPDIR Specify a temporary directory which replaces the default MAPRED.PID directory. –options=SCHEDOPTIONS If you want to add additional scheduler options, define them with –options as a single string.