lspvmrm - a PVM resource manager using load sharing facilities provided by LSF
lspvmrm [ -pvm options ] &
lspvmrm is an external PVM resource manager based upon the SM Protocol introduced in PVM 3.3. It interfaces with the LSF load sharing facility to provide dynamic load-balancing when scheduling PVM tasks onto to the hosts of the virtual machine.
Before running a pvm task lspvmrm should be started first. lspvmrm will start the PVM system if it isn't already running. Command line arguements to lspvmrm are ignored and are passed directly to the PVM system (see pvmd.1 and pvm.1) Once started the PVM console can be used to add other hosts to the virtual machine or to start application programs. Note that currently lspvmrm must be registered before any hosts are added to the system, otherwise the hosts won't know that lspvmrm exists.
lspvmrm logs information about it's actions to /tmp/pvmrm.<uid>.
lspvmrm utilizes the placement facilities of LSF to determine where best to schedule tasks, and if requested, which hosts to add to the Parallel Virtual Machine. The task scheduling feature is utilized automatically, but the selection of hosts must be requested explicitly.
When adding hosts the virtual machine, a host name is typically required. With lspvmrm, explicit host names can be replaced by host specifications. A host specifiction is made up of 3 parts in the following format: @[num_hosts:][LSF resource requirement strings]. The `@' is a flag character indicating that this is a host specification rather than a hostname. The `num_hosts' indicates the number of hosts that should be added (default is 1). A LSF resource requirement string can be used to specify the characteristics of the hosts to be added (see lsfintro.1).
For example when using the PVM console to add hosts, entering «add
@4:swap=25» will add 4 hosts with at least 25 megabytes of swap space. If
the pvm_addhosts(3PVM)
call is used, a host specfication may be used in place of the host name.
When using a host specification with the PVM console add command, error messages such as the following will be displayed:
pvm> add @4:cpu:swap=50
libpvm [t40002]: pvm_addhosts() sent count 1 received count 4
libpvm [t40002]: pvm_addhosts(): Out of resources
These errors should be ignored. Use the conf command to verify which hosts have been added.
When using the pvm_addhosts(3PVM) call with a host specification, the array of hosts should have elements to hold the number of hosts being requested. The first element should be a host specification. The remaining entries should be set to the placeholder specification «@0» as shown below:
char hosts[4] = {"@4:labfarm", «@0", «@0", «@0"}; int tids[4];
if (pvm_addhosts(hosts, 4, tids) < 0) /* works */
{
/* error handling code */
}
LSF specific error messages will not be visible from the console or returned by PVM API calls. These error messages will be logged in /tmp/pvmrm.<uid>.
LSF resource requirements cannot be used to delete heavily loaded hosts.
pvm(1PVM) , pvm_intro(1PVM) pvm_addhosts(3PVM) lsfintro(1)