DESCRIPTIONPVM is a software system that enables a collection of heterogeneous computers to be used as a coherent and flexible concurrent computational resource.
The individual computers may be shared- or local-memory multiprocessors, vector supercomputers, specialized graphics engines, or scalar workstations, that may be interconnected by a variety of networks, such as ethernet, FDDI.
User programs written in C, C++ or Fortran access PVM through library routines (libpvm3.a and libfpvm3.a).
Daemon programs (pvmd3) provide communication and process control between computers.
MACHINE ARCHITECTUREIn the PVM system, machines are assigned a short string to identify their architectures (this includes operating system type as well as CPU type). The types currently predefined in the distribution are:
- Alliant FX/8
- DEC Alpha/OSF-1
- DEC Alpha/OSF-1 / using shared memory
- IBM/RS6000 / AIX 4.x
- IBM SMP / shared memory transport / AIX 4.x
- IBM SP-2 / using MPI / AIX 4.x
- HP 300 running Domain/OS
- AT&T/NCR 3600 running SysVR4
- Sequent Balance
- BBN Butterfly TC2000
- 8086 running BSDI or BSD386
- Thinking Machines CM-2 Sun front-end
- Thinking Machines CM-5
- Convex using IEEE floating-point
- Convex using native f.p.
- Cray S-MP
- Convex Exemplar
- POSIX emulation layer on top of Windows32
- Data General Aviion
- Encore 88000
- 8086 running FreeBSD
- HP 9000 68000 cpu
- HP 9000 PA-Risc
- HP 9000 PA-Risc / shared memory transport
- Kendall Square
- Intel RX Hypercube
- Intel IPSC/2
- 8086 running Linux
- DEC Alpha running Linux
- Strogarm running Linux
- HP 9000 running Linux
- PowerPC running Linux
- Sparc running Linux
- Motorola M88100 running Real/IX
- Motorola M88100 running Real/IX
- DEC Alpha running NetBSD
- Amiga running NetBSD
- Strongarm running NetBSD
- HP 300 running NetBSD
- 8086 running NetBSD
- Any Motorola 68K running NetBSD
- Macintosh running NetBSD
- Mips EB running NetBSD
- Mips EL running NetBSD
- NS32K running NetBSD
- DEC Pmax running NetBSD
- PowerPC running NetBSD
- SH3 running NetBSD
- Sparc running NetBSD
- Sparc64 running NetBSD
- SUN 3 running NetBSD
- Vax running NetBSD
- Intel Paragon
- DEC/Mips arch (3100, 5000, etc.)
- IBM/RS6000 / AIX 3.x
- IBM SMP / shared memory transport / AIX 3.x
- 8086 running SCO Unix
- Silicon Graphics IRIS
- Silicon Graphics IRIS running OS 5.0
- Silicon Graphics IRIS running OS >= 6.0
- Silicon Graphics IRIS running 64 bit
- Silicon Graphics IRIS / OS 5.x / using shared memory
- Silicon Graphics IRIS / OS 6.x / using shared memory
- Silicon Graphics IRIS / 64 bit / using shared memory
- IBM SP-2 / using MPI / AIX 3.x
- Sun 3
- Sun 4, 4c, sparc, etc.
- Sun 4 running Solaris 2.x
- Sun 4 / using shared memory / Solaris 2.x
- NEC SX-3
- Sequent Symmetry
- Stardent Titan
- IBM 3090 running AIX
- Amdahl running UTS
- Fujitsu running UXP/M
- Thinking Machines CM-2 Vax front-end
- Windows 95/98/NT
- 8086 running Solaris 2.x
ENVIRONMENT VARIABLESThe following environment variables are read by PVM and may be set in order to customize your PVM environment. To set them, you can add commands to your .cshrc or .profile or equivalent shell startup file. See the manual page for the shell you normally use for information about how to do this. You can also include an appropriate shell startup file stub to set PVM environment variables and to add PVM directories to your execution path. Inert the matching stub file, pvm3/lib/cshrc.stub, pvm3/lib/kshrc.stub or pvm3/lib/bashrc.stub, after your declaration of PVM_ROOT in your shell startup file.
For csh users: Note that setting them in .login does not have the same effect. The .login script file is only read when you are actually logging in, whereas .cshrc is read every time csh starts up. PVM needs to have environment variables set when it starts a slave pvmd with "rsh host pvmd ...", so they must be set in .cshrc.
For those using a shell that doesn't always read a startup script (e.g. sh, ksh), there is another way to set environment variables for PVM. Before running the PVM executables, the pvm and pvmd startup scripts source any commands in $HOME/.pvmprofile if this file exists.
The following environment variables are supported by PVM 3.4.4:
- The path where PVM libraries and system programs are installed, for example /usr/local/pvm3 or $HOME/pvm3. This variable must be set on each host where PVM is used in order for PVM to function. There is no default value.
- The path for PVM temporary files, such as the daemon socket file pvmd.<uid> and the log file pvml.<uid>. Use this environment variable to use a directory other than /tmp (or C:\TEMP on Win32), or to introduce added security by using a protected subdirectory in /tmp that is owned by your userid and cannot be easily corrupted.
- The path to the "rsh" program on your system, if different than that defined in the $PVM_ROOT/conf/$PVM_ARCH.def configuration file. This environment variable can also be used to replace "rsh" with "ssh" for added security.
- The execution path to be searched for PVM programs on your system. By default, PVM looks in $HOME/pvm3/bin/$PVM_ARCH and $PVM_ROOT/bin/$PVM_ARCH for your PVM applications. This environment variable does not override the ep= host file option.
- The working directory for spawned PVM programs on your system. By default, PVM spawns your PVM applications in $HOME, but for convenience in accessing data or input files using relative path names, an alternate working directory can be specified. This environment variable does not override the wd= host file option.
- Names of environment variables to export from a parent task to children tasks through pvm_spawn(). Multiple names must be separated by ':'. If PVM_EXPORT is not set, no environment is exported.
- The debugger script to use when pvm_spawn() is called with PvmTaskDebug set. The default is $PVM_ROOT/lib/debugger.
The path of the pvmd startup script (default is $PVM_ROOT/lib/pvmd).
It is overridden by host file option dx=.
This variable is useful if you use a shell that doesn't automatically execute a startup script (such as .cshrc) to allow setting PVM_ROOT on slave (added) hosts. If you set it to the absolute or relative path of the pvmd startup script (for example /usr/local/pvm3/lib/pvmd or pvm3/lib/pvmd), the script will automatically set PVM_ROOT. Note that for this to work, you must set it to run the pvmd script, not the pvmd3 executable itself.
- This variable serves the same purpose as the PVM_DPATH above, but specifically for Win32 systems. This second environment variable allows for alternate specification of the path to the pvmd3.exe daemon executable using appropriate DOS file path syntax and environment variables (e.g. %PVM_ROOT%\lib\WIN32\pvmd3.exe).
- Specifies the path to an optional host file to be used by default when starting PVM. This alleviates the need to manually pass a host file path argument to the "pvm" console or "pvmd" script when starting PVM.
- Sets the maximum length of the pvmd error log file. Default value is the PVMDLOGMAX parameter in the source, 1 Mbyte.
- Sets the default pvmd debugging mask (as does the pvmd -d option). Value can be in hexadecimal (0x...), octal (0...) or decimal. Used to debug the pvmd (not intended to be used to debug application programs).
- Sets the default libpvm debugging mask (as does pvm_setopt(PvmDebugMask, x)). Value can be in hexadecimal (0x...), octal (0...) or decimal. Used to debug libpvm (not intended to be used to debug application programs).
- Sets additional flag bits for the pvm_spawn() library call. Allows override at run time of flags compiled into the pvm_spawn() calls in PVM application, e.g. to turn on PvmTaskDebug for popping up child tasks in a debugger window.
- Sets the size of the shared memory buffers used by libpvm and the pvmd. The default value is 1048576. If your program composes messages longer than this size, you must increase it.
A new feature in PVM 3.4.4 is the concept of a "Virtual Machine ID".
You can now set the PVM_VMID environment variable to an arbitrary
string (or use the "id=" option in a host file, see man page for pvmd3),
and this will distinguish and allow multiple virtual machines to run
on the same set of hosts under the same userid. (This feature was
originally introduced by SGI in their commercial PVM product, and has
now been generalized for the public PVM system.) This feature seems
to be something that people often want, and the PVM_VMID is the
cleanest way to provide this functionality, rather than overloading
the SHAREDTMP compiler flag and other internals.
Setting the PVM_VMID environment variable before starting PVM will create an encapsulated virtual machine with the given VMID name. By default, all other hosts which are added to this virtual machine will inherit the same VMID. If hosts are added to the virtual machine which are running older versions of PVM (prior to 3.4.4), then the VMID will be ignored for those hosts, and hence these machines can only be added to one virtual machine for the given user. The VMID need not be consistent on every host in a virtual machine (although this is not necessarily advisable), and the VMID can be set for individual hosts using the "id=" host file option (see man page for pvmd3).
The following environment variables are used by PVM internally. With the exception of PVM_ARCH, their values should not be modified. This is for information only.
- The PVM architecture name of the host on which it is set, used to distinguish between machines with different executable (a.out) formats. Copies of a program for different architectures are installed in parallel directories named for PVM architectures.
- Is passed from pvmd to spawned task, and gives the address of the pvmd local socket.
- Holds the expected process id of a spawned task exec'd by the pvmd. This is a magic cookie used by the task to identify itself when reconnecting to the pvmd, in order to get the correct task slot.
- The libpvm trace mask, passed from the pvmd to spawned tasks.
- The libpvm trace buffer size. If specified determines the number of bytes of trace event message buffer to be collected before sending to front-end tracer program.
- The libpvm trace option setting. Determines the level of tracing to be performed on invocations of PVM library calls.
- Used to optimize sending of PvmDataInPlace messages on MPP systems.
- PVM uses this value, combined with the process id, to generate shared-memory segment keys. The default value is your numeric uid. PVM automatically detects collisions when generating a key and picks a new key, so it should almost never need to be set explicitly.
AUTHORSA. L. Beguelin [4,5], J. J. Dongarra [1,2], G. A. Geist , W. C. Jiang , R. J. Manchek , B. K. Moore , V. S. Sunderam 
1. University of Tennessee, Knoxville TN.
2. Oak Ridge National Laboratory, Oak Ridge TN.
3. Emory University, Atlanta GA.
4. Carnegie Mellon University, Pittsburgh PA
5. Pittsburgh Supercomputer Center, Pittsburgh PA