NAME¶

pegasus-mpi-cluster - a tool for running computational workflows expressed as DAGs (Directed Acyclic Graphs) on computational clusters using MPI.

SYNOPSIS¶

pegasus-mpi-cluster [options] workflow.dag

DESCRIPTION¶

pegasus-mpi-cluster is a tool used to run HTC (High Throughput Computing) scientific workflows on systems designed for HPC (High Performance Computing). Many HPC systems have custom architectures that are optimized for tightly-coupled, parallel applications. These systems commonly have exotic, low-latency networks that are designed for passing short messages very quickly between compute nodes. Many of these networks are so highly optimized that the compute nodes do not even support a TCP/IP stack. This makes it impossible to run HTC applications using software that was designed for commodity clusters, such as Condor. pegasus-mpi-cluster was developed to enable loosely-coupled HTC applications such as scientific workflows to take advantage of HPC systems. In order to get around the network issues outlined above, pegasus-mpi-cluster uses MPI (Message Passing Interface), a commonly used API for writing SPMD (Single Process, Multiple Data) parallel applications. Most HPC systems have an MPI implementation that works on whatever exotic network architecture the system uses. An pegasus-mpi-cluster job consists of a single master process (this process is rank 0 in MPI parlance) and several worker processes. The master process manages the workflow and assigns workflow tasks to workers for execution. The workers execute the tasks and return the results to the master. Any output written to stdout or stderr by the tasks is captured (see TASK STDIO). pegasus-mpi-cluster applications are expressed as DAGs (Directed Acyclic Graphs) (see DAG FILES). Each node in the graph represents a task, and the edges represent dependencies between the tasks that constrain the order in which the tasks are executed. Each task is a program and a set of parameters that need to be run (i.e. a command and some optional arguments). The dependencies typically represent data flow dependencies in the application, where the output files produced by one task are needed as inputs for another. If an error occurs while executing a DAG that causes the workflow to stop, it can be restarted using a rescue file, which records the progress of the workflow (see RESCUE FILES). This enables pegasus-mpi-cluster to pick up running the workflow where it stopped. pegasus-mpi-cluster was designed to work either as a standalone tool or as a complement to the Pegasus Workflow Managment System (WMS). For more information about using PMC with Pegasus see the section on PMC AND PEGASUS. pegasus-mpi-cluster allows applications expressed as a DAG to be executed in parallel on a large number of compute nodes. It is designed to be simple, lightweight and robust.

OPTIONS¶

-h, --help -V, --version -v, --verbose -q, --quiet -s, --skip-rescue -o path, --stdout path -e path, --stderr path -m M, --max-failures M -t T, --tries T -n, --nolock -r, --rescue path --host-script path --host-memory size --host-cpus cpus --strict-limits --max-wall-time minutes --per-task-stdio --jobstate-log --monitord-hack --no-resource-log --no-sleep-on-recv --maxfds --keep-affinity

DAG FILES¶

pegasus-mpi-cluster workflows are expressed using a simple text-based format similar to that used by Condor DAGMan. There are only two record types allowed in a DAG file: TASK and EDGE. Any blank lines in the DAG (lines with all whitespace characters) are ignored, as are any lines beginning with # (note that # can only appear at the beginning of a line, not in the middle). The format of a TASK record is: Where id is the ID of the task, options is a list of task options, executable is the path to the executable or script to run, and arguments... is a space-separated list of arguments to pass to the task. An example is: This example specifies a task t01 that requires 10 MB memory and 2 CPUs to run /bin/program with the arguments -a and -b. The available task options are: -m M, --request-memory M -c N, --request-cpus N -t T, --tries T -p P, --priority P -f VAR=FILE, --pipe-forward VAR=FILE -F SRC=DEST, --file-forward SRC=DEST The format of an EDGE record is: Where parent is the ID of the parent task, and child is the ID of the child task. An example EDGE record is: A simple diamond-shaped workflow would look like this:

RESCUE FILES¶

Many different types of errors can occur when running a DAG. One or more of the tasks may fail, the MPI job may run out of wall time, pegasus-mpi-cluster may segfault (we hope not), the system may crash, etc. In order to ensure that the DAG does not need to be restarted from the beginning after an error, pegasus-mpi-cluster generates a rescue file for each workflow. The rescue file is a simple text file that lists all of the tasks in the workflow that have finished successfully. This file is updated each time a task finishes, and is flushed periodically so that if the work- flow fails and the user restarts it, pegasus-mpi-cluster can determine which tasks still need to be executed. As such, the rescue file is a sort-of transaction log for the workflow. The rescue file contains zero or more DONE records. The format of these records is: Where taskid is the ID of the task that finished successfully. By default, rescue files are named DAGNAME.rescue where DAGNAME is the path to the input DAG file. The file name can be changed by specifying the -r argument.

PMC AND PEGASUS¶

Using PMC for Pegasus Task Clustering¶

PMC can be used as the wrapper for executing clustered jobs in Pegasus. In this mode Pegasus groups several tasks together and submits them as a single clustered job to a remote system. PMC then executes the individual tasks in the cluster and returns the results. PMC can be specified as the task manager for clustered jobs in Pegasus in three ways: It is usually necessary to have a pegasus::mpiexec entry in your transformation catalog that specifies a) the path to PMC on the remote site and b) the relevant globus profiles such as xcount, host_xcount and maxwalltime to control size of the MPI job. That entry would look like this: If this transformation catalog entry is not specified, Pegasus will attempt create a default path on the basis of the environment profile PEGASUS_HOME specified in the site catalog for the remote site. PMC can be used with both horizontal and label-based clustering in Pegasus, but we recommend using label-based clustering so that entire sub-graphs of a Pegasus DAX can be clustered into a single PMC job, instead of only a single level of the workflow.

Pegasus Profiles for PMC¶

There are several Pegasus profiles that map to PMC task options: pmc_request_memory pmc_request_cpus pmc_priority These profiles are used by Pegasus when generating PMC’s input DAG when PMC is used as the task manager for clustered jobs in Pegasus. The profiles can be specified in the DAX like this: This example specifies a PMC task that requires 1GB of memory and 4 cores, and has a priority of 10. It produces a task in the PMC DAG that looks like this:

Using PMC for the Entire Pegasus DAX¶

Pegasus can also be configured to run the entire workflow as a single PMC job. In this mode Pegasus will generate a single PMC DAG for the entire workflow as well as a PBS script that can be used to submit the workflow. In contrast to using PMC as a task clustering tool, in this mode there are no jobs in the workflow executed without PMC. The entire workflow, including auxilliary jobs such as directory creation and file transfers, is managed by PMC. If Pegasus is configured in this mode, then DAGMan and Condor are not required. To run in PMC-only mode, set the property "pegasus.code.generator" to "PMC" in the Pegasus properties file: In order to submit the resulting PBS job you may need to make changes to the .pbs file generated by Pegasus to get it to work with your cluster. This mode is experimental and has not been used extensively.

LOGGING¶

By default, all logging messages are printed to stderr. If you turn up the logging using -v then you may end up with a lot of stderr being forwarded from the workers to the master. The log levels in order of severity are: FATAL, ERROR, WARN, INFO, DEBUG, and TRACE. The default logging level is INFO. The logging levels can be increased with -v and decreased with -q.

TASK STDIO¶

By default the stdout and stderr of tasks will be redirected to the master’s stdout and stderr. You can change the path of these files with the -o and -e arguments. You can also enable per-task stdio files using the --per-task-stdio argument. Note that if per-task stdio files are not used then the stdio of all workers will be merged into one out and one err file by the master at the end, so I/O from different workers will not be interleaved, but I/O from each worker will appear in the order that it was generated. Also note that, if the job fails for any reason, the outputs will not be merged, but instead there will be one file for each worker named DAGFILE.out.X and DAGFILE.err.X, where DAGFILE is the path to the input DAG, and X is the worker’s rank.

HOST SCRIPTS¶

A host script is a shell script or executable that pegasus-mpi-cluster launches on each unique host on which it is running. They can be used to start auxilliary services, such as memcached, that the tasks in a workflow require. Host scripts are specified using either the --host-script argument or the PMC_HOST_SCRIPT environment variable. The host script is started when pegasus-mpi-cluster starts and must exit with an exitcode of 0 before any tasks can be executed. If it the host script returns a non-zero exitcode, then the workflow is aborted. The host script is given 60 seconds to do any setup that is required. If it doesn’t exit in 60 seconds then a SIGALRM signal is delivered to the process, which, if not handled, will cause the process to terminate. When the workflow finishes, pegasus-mpi-cluster will deliver a SIGTERM signal to the host script’s process group. Any child processes left running by the host script will receive this signal unless they created their own process group. If there were any processes left to receive this signal, then they will be given a few seconds to exit, then they will be sent SIGKILL. This is the mechanism by which processes started by the host script can be informed of the termination of the workflow.

RESOURCE-BASED SCHEDULING¶

High-performance computing resources often have a low ratio of memory to CPUs. At the same time, workflow tasks often have high memory requirements. Often, the memory requirements of a workflow task exceed the amount of memory available to each CPU on a given host. As a result, it may be necessary to disable some CPUs in order to free up enough memory to run the tasks. Similarly, many codes have support for multicore hosts. In that case it is necessary for efficiency to ensure that the number of cores required by the tasks running on a host do not exceed the number of cores available on that host. In order to make this process more efficient, pegasus-mpi-cluster supports resource-based scheduling. In resource-based scheduling the tasks in the workflow can specify how much memory and how many CPUs they require, and pegasus-mpi-cluster will schedule them so that the tasks running on a given host do not exceed the amount of physical memory and CPUs available. This enables pegasus-mpi-cluster to take advantage of all the CPUs available when the tasks' memory requirement is low, but also disable some CPUs when the tasks' memory requirement is higher. It also enables workflows with a mixture of single core and multi-core tasks to be executed on a heterogenous pool. If there are no hosts available that have enough memory and CPUs to execute one of the tasks in a workflow, then the workflow is aborted.

Memory¶

Users can specify both the amount of memory required per task, and the amount of memory available per host. If the amount of memory required by any task exceeds the available memory of all the hosts, then the workflow will be aborted. By default, the host memory is determined automatically, however the user can specify --host-memory to "lie" to pegasus-mpi-cluster. The amount of memory required for each task is specified in the DAG using the -m/--request-memory argument (see DAG Files).

CPUs¶

Users can specify the number of CPUs required per task, and the total number of CPUs available on each host. If the number of CPUs required by a task exceeds the available CPUs on all hosts, then the workflow will be aborted. By default, the number of CPUs on a host is determined automatically, but the user can specify --host-cpus to over- or under-subscribe the host. The number of CPUs required for each task is specified in the DAG using the -c/ --request-cpus argument (see DAG Files).

I/O FORWARDING¶

In workflows that have lots of small tasks it is common for the I/O written by those tasks to be very small. For example, a workflow may have 10,000 tasks that each write a few KB of data. Typically each task writes to its own file, resulting in 10,000 files. This I/O pattern is very inefficient on many parallel file systems because it requires the file system to handle a large number of metadata operations, which are a bottleneck in many parallel file systems. One way to handle this problem is to have all 10,000 tasks write to a single file. The problem with this approach is that it requires those tasks to synchronize their access to the file using POSIX locks or some other mutual exclusion mechanism. Otherwise, the writes from different tasks may be interleaved in arbitrary order, resulting in unusable data. In order to address this use case PMC implements a feature that we call "I/O Forwarding". I/O forwarding enables each task in a PMC job to write data to an arbitrary number of shared files in a safe way. It does this by having PMC worker processes collect data written by the task and send it over over the high-speed network using MPI messaging to the PMC master process, where it is written to the output file. By having one process (the PMC master process) write to the file all of the I/O from many parallel tasks can be synchronized and written out to the files safely. There are two different ways to use I/O forwarding in PMC: pipes and files. Pipes are more efficient, but files are easier to use.

I/O forwarding using pipes¶

I/O forwarding with pipes works by having PMC worker processes collect data from each task using UNIX pipes. This approach is more efficient than the file-based approach, but it requires the code of the task to be changed so that the task writes to the pipe instead of a regular file. In order to use I/O forwarding a PMC task just needs to specify the -f/--pipe-forward argument to specify the name of the file to forward data to, and the name of an environment variable through which the PMC worker process can inform it of the file descriptor for the pipe. For example, if there is a task "mytask" that needs to forward data to two files: "myfile.a" and "myfile.b", it would look like this: When the /bin/mytask process starts it will have two variables in its environment: "A=3" and "B=4", for example. The value of these variables is the file descriptor number of the corresponding files. In this case, if the task wants to write to "/tmp/myfile.a", it gets the value of environment variable "A", and calls write() on that descriptor number. In C the code for that looks like this: In some programming languages it is not possible to write to a file descriptor directly. Fortran, for example, refers to files by unit number instead of using file descriptors. In these languages you can either link C I/O functions into your binary and call them from routines written in the other language, or you can open a special file in the Linux /proc file system to get another handle to the pipe you want to access. For the latter, the file you should open is "/proc/self/fd/NUMBER" where NUMBER is the file descriptor number you got from the environment variable. For the example above, the pipe for myfile.a (environment variable A) is "/proc/self/fd/3". If you are using pegasus-kickstart, which is probably the case if you are using PMC for a Pegasus workflow, then there’s a trick you can do to avoid modifying your code. You use the /proc file system, as described above, but you let pegasus-kickstart handle the path construction. For example, if your application has an argument, -o, that allows you to specify the output file then you can write your task like this: In this case, pegasus-kickstart will replace the $A in your application arguments with the file descriptor number you want. Your code can open that path normally, write to it, and then close it as if it were a regular file.

I/O forwarding using files¶

I/O forwarding with files works by having tasks write out data in files on the local disk. The PMC worker process reads these files and forwards the data to the master where it can be written to the desired output file. This approach may be much less efficient than using pipes because it involves the file system, which has more overhead than a pipe. File forwarding can be enabled by giving the -F/--file-forward argument to a task. Here’s an example: In this case, the worker process will expect to find the file /tmp/foo.0 when mytask exits successfully. It reads the data from that file and sends it to the master to be written to the end of /scratch/foo. After /tmp/foo.0 is read it will be deleted by the worker process. This approach works best on systems where the local disk is a RAM file system such as Cray XT machines. Alternatively, the task can use /dev/shm on a regular Linux cluster. It might also work relatively efficiently on a local disk if the file system cache is able to absorb all of the reads and writes.

I/O forwarding caveats¶

When using I/O forwarding it is important to consider a few caveats. First, if the PMC job fails for any reason (including when the workflow is aborted for violating --max-wall-time), then the files containing forwarded I/O may be corrupted. They can include partial records, meaning that only part of the I/O from one or more tasks was written, and they can include duplicate records, meaning that the I/O was written, but the PMC job failed before the task could be marked as successful, and the workflow was restarted later. We make no guarantees about the contents of the data files in this case. It is up to the code that reads the files to a) detect and b) recover from such problems. To eliminate duplicates the records should include a unique identifier, and to eliminate partials the records should include a checksum. Second, you should not use I/O forwarding if your task is going to write a lot of data to the file. Because the PMC worker is reading data off the pipe/file into memory and sending it in an MPI message, if you write too much, then the worker process will run the system out of memory. Also, all the data needs to fit in a single MPI message. In pipe forwarding there is no hard limit on the size, but in file forwarding the limit is 1MB. We haven’t benchmarked the performance on large I/O, but anything larger than about 1 MB is probably too much. At any rate, if your data is larger than 1MB, then I/O forwarding probably won’t have much of a performance benefit anyway. Third, the I/O is not written to the file if the task returns a non-zero exitcode. We assume that if the task failed that you don’t want the data it produced. Fourth, the data from different tasks is not interleaved. All of the data written by a given task will appear sequentially in the output file. Note that you can still get partial records, however, if any data from a task appears it will never be split among non-adjacent ranges in the output file. If you have 3 tasks that write: "I am a task" you can get: and: but not: Fifth, data from different tasks appears in arbitrary order in the output file. It depends on what order the tasks were executed by PMC, which may be arbitrary if there are no dependencies between the tasks. The data that is written should contain enough information that you are able to determine which task produced it if you require that. PMC does not add any headers or trailers to the data. Sixth, a task will only be marked as successful if all of its I/O was successfully written. If the workflow completed successfully, then the I/O is guaranteed to have been written. Seventh, if the master is not able to write to the output file for any reason (e.g. the master tries to write the I/O to the destination file, but the write() call returns an error) then the task is marked as failed even if the task produced a non-zero exitcode. In other words, you may get a non-zero kickstart record even when PMC marks the task failed. Eighth, the pipes are write-only. If you need to read and write data from the file you should use file forwarding and not pipe forwarding. Ninth, all files are opened by the master in append mode. This is so that, if the workflow fails and has to be restarted, or if a task fails and is retried, the data that was written previously is not lost. PMC never truncates the files. This is one of the reasons why you can have partial records and duplicate records in the output file. Finally, in file forwarding the output file is removed when the task exits. You cannot rely on the file to be there when the next task runs even if you write it to a shared file system.

MISC¶

Resource Utilization¶

At the end of the workflow run, the master will report the resource utilization of the job. This is done by adding up the total runtimes of all the tasks executed (including failed tasks) and dividing by the total wall time of the job times N, where N is both the total number of processes including the master, and the total number of workers. These two resource utilization values are provided so that users can get an idea about how efficiently they are making use of the resources they allocated. Low resource utilization values suggest that the user should use fewer cores, and longer wall time, on future runs, while high resource utilization values suggest that the user could use more cores for future runs and get a shorter wall time.

KNOWN ISSUES¶

fork() and exec()¶

In order for the worker processes to start tasks on the compute node the compute nodes must support the fork() and exec() system calls. If your target machine runs a stripped-down OS on the compute nodes that does not support these system calls, then pegasus-mpi-cluster will not work.

CPU Usage¶

Many MPI implementations are optimized so that message sends and receives do busy waiting (i.e. they spin/poll on a message send or receive instead of sleeping). The reasoning is that sleeping adds overhead and, since many HPC systems use space sharing on dedicated hardware, there are no other processes competing, so spinning instead of sleeping can produce better performance. On those implementations MPI processes will run at 100% CPU usage even when they are just waiting for a message. This is a big problem for multicore tasks in pegasus-mpi-cluster because idle slots consume CPU resources. In order to solve this problem pegasus-mpi-cluster processes sleep for a short period between checks for waiting messages. This reduces the load significantly, but causes a short delay in receiving messages. If you are using an MPI implementation that sleeps on message send and receive instead of doing busy waiting, then you can disable the sleep by specifying the --no-sleep-on-recv option. Note that the master will always sleep if --max-wall-time is specified because there is no way to interrupt or otherwise timeout a blocking call in MPI (e.g. SIGALRM does not cause MPI_Recv to return EINTR).

ENVIRONMENT VARIABLES¶

The environment variables below are aliases for command-line options. If the environment variable is present, then it is used as the default for the associated option. If both are present, then the command-line option is used. PMC_HOST_SCRIPT PMC_HOST_MEMORY PMC_HOST_CPUS PMC_MAX_WALL_TIME

AUTHOR¶

Gideon Juve <gideon@isi.edu> Mats Rynge <rynge@isi.edu>