NAME

jsd —

simple command scheduling daemon for remote execution

SYNOPSIS

jsd [-iqv] [-f fanout] [-g rungroup1,...,rungroupN] [-l username] [-p port] [-o port] [-w node1,...,nodeN] [-x node1,...,nodeN] [command ...]

DESCRIPTION

The jsd daemon is used to schedule remote execution of jobs across a heterogenous cluster of machines. Upon starting the daemon, it will attempt to run a benchmark on all machines under it's control. Once the daemon is active it listens for connections from a jsh process, and gives each connecting process the name of a remote host to execute a command on. Faster hosts are preferred by the jsd daemon, and the fastest of all remaining hosts will allways be given out first. Once a remote jsh process has completed it's use of the remote host, it will again contact the jsd daemon, and return the node to service. The daemon will never give the same node out to more than one process at a time.

jsd logs errors and critical conditions to the syslog with the facility LOG_DAEMON.

The following options are available:

-i: The -i option will list information about the current cluster, and command groupings. It will print out the current value of the fanout, and how many groups of machines there are within the cluster. It will also show you which command you are about to run, and your username if specified with the -l option.
-q: The -q option does not issue any commands, but displays information about the cluster, and the fanout groupings.
-v: Prints the version of ClusterIt to the stdout, and exits.
-f: If the -f option is specified, followed by a number, it sets the fanout size of the cluster. The fanout size is the number of nodes a command will run on in parallel at one time. Thus a 80 node cluster, with a fanout size of 64, would run 64 nodes in parallel, then, when all have finished, it would execute the command on the last 16 nodes. The fanout size defaults to 64. This option overrides the FANOUT environment variable.
-g: If the -g option is specified, followed by a comma separated list of group names, the command will only be run on that group of nodes. A node may be a part of more than one group if desired, however running without the -g option will run the command on the same node as many times as it appears in the file specified by the CLUSTER environment variable. This option is silently ignored if used with the -w option.
-l: If the -l option is specified, followed by a username, the commands will be run under that userid on the remote machines. Consideration must be taken for proper authentication, for this to work.
-p: The -p option can be used to set the port number which jsd will listen to replies from jsh processes indicating completion of processing on remote nodes. This option overrides the JSD_IPORT environment variable. The default listen port is 2001.
-o: The -o option can be used to set the port number which jsd will listen to requests from jsh processes for remote nodes. This option overrides the JSD_OPORT environment variable. The default listen port is 2002.
-w: If the -w option is specified, followed by a comma delimited list of machine names, the command will be run on each node in the list. Without this option, jsd runs on the nodes listed in the file pointed to by the CLUSTER environment variable.
-x: The -x option can be used to exclude specific nodes from the cluster. The format is the same as the -w option, a comma delimited list of machine names. This option is silently ignored if used with the -w option.

ENVIRONMENT

jsd utilizes the following environment variables:

CLUSTER: Contains a filename, which is a newline separated list of nodes in the cluster.
RCMD_CMD: Command to use to connect to remote machines. The command chosen must be able to connect with no password to the remote host. Defaults to rsh. This is only used by jsd when issuing the benchmark at startup, and has no effect on how jsh processes will connect to the remote nodes.
RCMD_CMD_ARGS: Arguments to pass to the remote shell command. Defaults to none. As with RCMD_CMD this is only used when issuing the benchmark at startup.
JSD_BENCH_CMD: This is the name of a command, which when run on the remote nodes of the cluster will give some indication of the relative speed of the machines. It does not need to be precise. If no benchmark command is specified, all machines are considered to be the same speed, and will not be wieghted. The bench command expects a single floating point number as the only output to the remote command. Larger numbers are to indicate faster machines. See the EXAMPLES section for more details.
FANOUT: When set, limits the maximum number of concurrent commands sent at once. This can be used to keep from overloading a small host when sending out commands in parallel. Defaults to 64. This environment setting can be overridden by the -f option.
JSD_OPORT: When set, overrides the default port which jsd listens on for jsh processes requesting nodes. This environment setting can be overriden by the -o option.
JSD_IPORT: When set, overrides the default port which jsd listens on for jsh processes reporting back that it has completed processing on a node. This environment setting can be overriden by the -p option.

EXAMPLES

In order to generate a benchmark figure useful to jsd you will most likely need to process the ouput of a benchmark command in some way. There are two simple ways of doing this:

The first way is to build a script, which exists on each machine, and run that as your benchmark command. The below script shows how to process the output of the whetstone(1) command into a format useable by jsd :

#!/bin/sh
 ./whetstone 100000 | grep MIPS | sed -e 's|.*: \(.*\) MIPS|\1|'

The other way, would be to run a more involved benchmark on each machine, (or assign numbers arbitrarily) and place them in a file on the remote machines. A benchmark command could then simply be cat /etc/benchmark or some other similar setup.

HISTORY

The jsd command first appeared in clusterit 2.0. It should be pronounced “jez-dee”, and stands for ‘Job Scheduling Daemon’.

AUTHOR

Jsd was written by Tim Rightnour.