I need to debug an MPI code for which I only have access to a single node/machine. The problem is the bug I am looking for only arises when running on more than node but it doesn't when running, for example, two MPI tasks in the same node, everything goes fine. I assume that my MPI implementation (mviapich2) cleverly treats tasks running on the same node by, for example, replacing network communications by IPC strategies or even memcpy.
So my question is: how could I run two MPI tasks on a single node but making MPI treat them as tasks on different nodes? Is that possible?
You can disable the MVAPICH2 shared memory device by setting the MV2_USE_SHARED_MEM environment variable to 0:
mpiexec ... -env MV2_USE_SHARED_MEM 0 ... ./executable
Make sure that your MVAPICH2 was built with the TCP/IP device, otherwise your ranks won't be able to communicate with shared memory support turned off.
Related
I am trying to implement an MPI program which is to have a server node assign task pieces to client node, but I am a freshman and don't know how to mange the client list, can anyone help me?
Let me describe it more specify:
Server node:
MPI_COMM clients[4]; // store client communicators, but I am not sure this is correct or not!
Several clients connect to server using mpirun -np 1 ./mpiclient more than one time, not processor greater than 1.
The reason I want to do this is I want to send each client different job to calculate.
Question 2: How can I get the attributes of the client?
For example: MPI_Comm_accept(portname,MPI_INFO_NULL,0,MPI_COMM_SELF,&client);
How can I get client name or ip?
I don't know that there's a best practice here, but there are a few options.
Know your list of IP addresses ahead of time.
Most of the time, people have a cluster set up with a static pool of IP addresses. That means its easy to predict who will be connecting so you can call MPI_COMM_ACCEPT for each IP address and the clients will already know the address of the "server".
Don't use Connect/Accept directly
It may not be necessary to go through the pain of managing all of your connections directly. You might be able to do something else like MPI_COMM_SPAWN(_MULTIPLE) and spawn your children directly from your master. This simplifies managing connections, though you still have to deal with some of the weirdness of dynamic processing in MPI, specifically inter-communciators.
Don't use dynamic processes at all
Many times, people coming into distributed programming, and specifically MPI, for the first time still have a sockets frame of reference. Meaning, they expect to have to set up all of their own connections and communication management. In reality, MPI and other communication libraries are designed to be slightly higher level than that and let you ignore some of the mundane communication management and get straight to passing data around. Usually in an MPI job, you will use a single binary for your program and have each process decide what they'll be doing based on their rank. For example:
mpirun -np 5 ./my_prog
...
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
if (rank == 0) {
/* Distribute work */
} else {
/* Get work from rank 0 */
}
...
It's also possible to run multiple binaries this way if you want to separate out your codes for different parts of the program. How you run this can vary from implementation to implementation, but with MPICH, it works like this:
mpiexec -n 1 ./my_prog1 : -n 4 ./my_prog2
Then my_prog1 would distribute the work to all of the other processes running my_prog2. In this model, all processes still end up in the same MPI_COMM_WORLD so they can just check their rank at the beginning of the program and get working.
When starting an MPI job with mpirun or mpiexec, I can understand how one might go about starting each individual process. However, without any compiler magic, how do these wrapper executables communicate the arrangement (MPI communicator) to the MPI processes?
I am interested in the details, or a pointer on where to look.
Details on how individual processes establish the MPI universe are implementation specific. You should look into the source code of the specific library in order to understand how it works. There are two almost universal approaches though:
command line arguments: the MPI launcher can pass arguments to the spawned processes indicating how and where to connect in order to establish the universe. That's why MPI has to be initialised by calling MPI_Init() with argc and argv in C - thus the library can get access to the command line and extract all arguments that are meant for it;
environment variables: the MPI launcher can set specific environment variables whose content can indicate where and how to connect.
Open MPI for example sets environment variables and also writes some universe state in a disk location known to all processes that run on the same node. You can easily see the special variables that its run-time component ORTE (OpenMPI Run-Time Environment) uses by executing a command like mpirun -np 1 printenv:
$ mpiexec -np 1 printenv | grep OMPI
... <many more> ...
OMPI_MCA_orte_hnp_uri=1660944384.0;tcp://x.y.z.t:43276;tcp://p.q.r.f:43276
OMPI_MCA_orte_local_daemon_uri=1660944384.1;tcp://x.y.z.t:36541
... <many more> ...
(IPs changed for security reasons)
Once a child process is launched remotely and MPI_Init() or MPI_Init_thread() is called, ORTE kicks in and reads those environment variables. Then it connects back to the specified network address with the "home" mpirun/mpiexec process which then coordinates all spawned processes into establishing the MPI universe.
Other MPI implementations work in a similar fashion.
My situation is quite simple: I want to run a MPI-enabled software on a single multiprocessor/core machine, let's say 8.
My implementation of MPI is MPICH2.
As I understand I have a few options:
$ mpiexec -n 8 my_software
$ mpiexec -n 8 -hosts {localhost:8} my_software
or I could also specify Hydra to "fork" and not "ssh";
$ mpiexec -n 8 -launcher fork my_software
Could you tell me if there will be any differences or if the behavior will be the same ?
Of course as all my nodes will be on the same machine I don't want "message passing" to be done through the network (even the local loop) but through shared memory. As I understood MPI will figure that out itself and that will be the case for all the three options.
Simple answer:
All methods should lead to the same performance. You'll have 8 processes running on the cores and using shared memory.
Technical answer:
"fork" has the advantage of compatibility, on systems where rsh/ssh process spawning would be a problem. But can, I guess, only start processes locally.
At the end (unless MPI is weirdly configured) all processes on the same CPU will end up using "shared memory", and the launcher or the host specification method should not matter for this. The communication method is handled by another parameter (-channel ?).
Specific syntax of host specification method can permit to bind processes to a specific CPU core, then you might have slightly better/worse performance depending of your application.
If you've got everything set up correctly then I don't see that your program's behaviour will depend on how you launch it, unless that is it fails to launch under one or other of the options. (Which would mean that you didn't have everything set up correctly in the first place.)
If memory serves me well the way in which message passing is implemented depends on the MPI device(s) you use. It used to be that you would use the mpi ch_shmem device. This managed the passing of messages between processes but it did use buffer space and messages were sent to and from this space. So message passing was done, but at memory bus speed.
I write in the past tense because it's a while since I was that close to the hardware that I knew (or, frankly, cared) about low-level implementation details and more modern MPI installations might be a bit, or a lot, more sophisticated. I'll be surprised, and pleased, to learn that any modern MPI installation does, in fact, replace message-passing with shared memory read/write on a multicore/multiprocessor machine. I'll be surprised because it would require translating message-passing into shared memory access and I'm not sure that that is easy (or easy enough to be feasible) for the whole of MPI. I think it's far more likely that current implementations still rely on message-passing across the memory bus through some buffer area. But, as I state, that's only my best guess and I'm often wrong on these matters.
Is it possible to hot plug an additional node (host) into a working OpenMPI app? We're talking about production environment where we cannot afford even a 5 second downtime.
There are two scenarios I'm interested in:
We just would like to enhance the computing power by adding one more broadcast listener.
A node died, the master node handles it well and reassigns the task to somebody else. The system administrator comes in, restarts the dead node and plugs it back into the cluster.
Which platform independent MPI implementation would be best for the scenario above? OpenMPI is not a must here.
MPI-2 -- any implementation -- does allow dynamic processes, and in fact adding processes is currently much more feasible than removing processes. You can use MPI_COMM_SPAWN to launch a new process with a given executable, and that returns an intracommunicator that can be used to communicate between the old (original) processes.
The tricks here are -- nothing will automatically detect the new node. You'll have to have some process keeping an eye out for them, SPAWN something on them. If the new nodes will just be listeners to the master node, that's probably the best case, as only the master node really needs to know about it. The invocation to ensure the spawn happens on the new node and not somewhere else will be done through the info argument to spawn, and may be implementation dependant.
I want to easily perform collective communications independently on each machine of my cluster. Let's say I have 4 machines with 8 cores on each, my MPI program would run 32 MPI tasks. What I would like is, for a given function:
on each host, only one task performs a computation, the other tasks do nothing during this computation. In my example, 4 MPI tasks will do the computation, 28 others are waiting.
once the computation is done, each MPI task on each will perform a collective communication ONLY to local tasks (tasks running on the same host).
Conceptually, I understand I must create one communicator for each host. I searched around, and found nothing explicitly doing that. I am not really comfortable with MPI groups and communicators. Here my two questions:
is MPI_Get_processor_name is enough unique for such a behaviour?
more generally, do you have a piece of code doing that?
The specification says that MPI_Get_processor_name returns "A unique specifier for the actual (as opposed to virtual) node", so I think you'd be ok with that. I guess you'd do a gather to assemble all the host names and then assign groups of processors to go off and make their communicators; or dup MPI_COMM_WORLD, turn the names into integer hashes, and use mpi_comm_split to partition the set.
You could also take the approach janneb suggests and use implementation-specific options to mpirun to ensure that the MPI implementation assigns tasks that way; OpenMPI uses --byslot to generate this ordering; with mpich2 you can use -print-rank-map to see the mapping.
But is this really what you want to do? If the other processes are sitting idle while one processor is working, how is this better than everyone redundantly doing the calculation? (Or is this very memory or I/O intensive, and you're worried about contention?) If you're going to be doing a lot of this -- treating on-node parallelization very different from off-node parallelization -- then you may want to think about hybrid programming models - running one MPI task per node and MPI_spawning subtasks or using OpenMP for on-node communications, both as suggested by HPM.
I don't think (educated thought, not definitive) that you'll be able to do what you want entirely from within your MPI program.
The response of the system to a call to MPI_Get_processor_name is system-dependent; on your system it might return node00, node01, node02, node03 as appropriate, or it might return my_big_computer for whatever processor you are actually running on. The former is more likely, but it is not guaranteed.
One strategy would be to start 32 processes and, if you can determine what node each is running on, partition your communicator into 4 groups, one on each node. This way you can manage inter- and intra-communications yourself as you wish.
Another strategy would be to start 4 processes and pin them to different nodes. How you pin processes to nodes (or processors) will depend on your MPI runtime and any job management system you might have, such as Grid Engine. This will probably involve setting environment variables -- but you don't tell us anything about your run-time system so we can't guess what they might be. You could then have each of the 4 processes dynamically spawn a further 7 (or 8) processes and pin those to the same node as the initial process. To do this, read up on the topic of intercommunicators and your run-time system's documentation.
A third strategy, now it's getting a little crazy, would be to start 4 separate MPI programs (8 processes each), one on each node of your cluster, and to join them as they execute. Read about MPI_Comm_connect and MPI_Open_port for details.
Finally, for extra fun, you might consider hybridising your program, running one MPI process on each node, and have each of those processes execute an OpenMP shared-memory (sub-)program.
Typically your MPI runtime environment can be controlled e.g. by environment variables how tasks are distributed over nodes. The default tends to be sequential allocation, that is, for your example with 32 tasks distributed over 4 8-core machines you'd have
machine 1: MPI ranks 0-7
machine 2: MPI ranks 8-15
machine 3: MPI ranks 16-23
machine 4: MPI ranks 24-31
And yes, MPI_Get_processor_name should get you the hostname so you can figure out where the boundaries between hosts are.
The modern MPI 3 answer to this is to call MPI_Comm_split_type