I've got an MPI program consisting of one master process that hands off commands to a bunch of slave processes. Upon receiving a command, a slave just calls system() to do it. While the slaves are waiting for a command, they are consuming 100% of their respective CPUs. It appears that Probe() is sitting in a tight loop, but that's only a guess. What do you think might be causing this, and what could I do to fix it?
Here's the code in the slave process that waits for a command. Watching the log and the top command at the same time suggests that when the slaves are consuming their CPUs, they are inside this function.
MpiMessage
Mpi::BlockingRecv() {
LOG(8, "BlockingRecv");
MpiMessage result;
MPI::Status status;
MPI::COMM_WORLD.Probe(MPI_ANY_SOURCE, MPI_ANY_TAG, status);
result.source = status.Get_source();
result.tag = status.Get_tag();
int num_elems = status.Get_count(MPI_CHAR);
char buf[num_elems + 1];
MPI::COMM_WORLD.Recv(
buf, num_elems, MPI_CHAR, result.source, result.tag
);
result.data = buf;
LOG(7, "BlockingRecv about to return (%d, %d)", result.source, result.tag);
return result;
}
Yes; most MPI implementations, for the sake of performance, busy-wait on blocking operations. The assumption is that the MPI job is the only thing going on that we care about on the processor, and if the task is blocked waiting for communications, the best thing to do is to continually poll for that communication to reduce latency; so that there's virtually no delay between when the message arrives and when it's handed off to the MPI task. This typically means that CPU is pegged at 100% even when nothing "real" is being done.
That's probably the best default behaviour for most MPI users, but it isn't always what you want. Typically MPI implementations allow turning this off; with OpenMPI, you can turn this behaviour off with an MCA parameter,
mpirun -np N --mca mpi_yield_when_idle 1 ./a.out
It sounds like there are three ways to wait for an MPI message:
Aggressive busy wait. This will get the message into your receiving code as fast as possible. Some processor is doing nothing but checking for the incoming message. If you put all of your processors in this state, the rest of your system is going to be very slow. MPI uses aggressive mode by default.
Degraded busy wait. This will yield to other processes while doing its busy wait. If the number of processes you ask for is more than the number of processors you have, MPI switches to degraded mode. You can also force aggressive or degraded mode with an MCA parameter.
Polling. Even the degraded busy wait is still a busy wait, and it will keep one processor pegged at 100% for each process that is waiting. If you have other tasks on your system that you don't want to compete with, you can call MPI_Iprobe() in a loop with a sleep call before calling a blocking receive. I find a 100ms sleep is responsive enough for my tasks, and still keeps the CPU usage minimal when a worker is idle.
I did some searching and found that a busy wait is what you want if you are not sharing your processors with other tasks.
Related
I am teaching myself OS by going through the lecture notes of the course at IIT Bombay (https://www.cse.iitb.ac.in/~mythili/os/). One of the questions in the Process worksheet asks which of the following doesn't always happen in the situation described at the title. The answer is C.
A. The process moves to kernel mode.
B. The program counter of the CPU shifts to the kernel part of the address space.
C. The process is context-switched out and a separate kernel process starts execution.
D. The OS code that deals with handling TCP/IP packets is invoked
I'm a bit confused though. I thought when an interrupt routine occurs the process is context-switched out so other processes can run and the CPU is not idle during that time. The kernel, then, will take care of the packet sending. Why would C not be correct then?
You are right in saying that "when an interrupt routine occurs the process is context-switched out so other processes can run and the CPU is not idle during that time", but the words "generally or mostly" need to be added to it.
In most cases, there is another process waiting for CPU time and that can be scheduled. However it is not the case 100% of the time. The question is about the word "always" and while other options always occur in the given situation, option C is a choice that OS makes at run time. If OS determines that switching out this process can be sub optimal than performing the system call and resuming the same process, then it may not perform the context switching.
There is a cost associated with context switching and if other processes are also blocked on some I/O then it may be optimal for OS to NOT switch the context or there might be other reasons to not switch the context such as what if only 1 process is running, there is no other process to switch the context to!
I'm new to OpenCL programming. In one of my OpenCL applications, I use clWaitForEvents after launching every kernel.
Case 1:
time_start();
cl_event event;
cl_int status = clEnqueueNDRangeKernel(queue, ..., &event);
clWaitForEvents(1, &event);
time_end();
Time taken : 250 ms (with clWaitForEvents)
If I remove clWaitForEvents(), my kernel runs faster with the same output.
Case 2:
time_start();
cl_event event;
cl_int status = clEnqueueNDRangeKernel(queue, ..., &event);
time_end();
Time taken: 220 ms (without clWaitForEvents)
I've to launch 10 different kernels sequentially. Every kernel is dependent on the output of the previous kernel. Using clWaitForEvent after every kernel increases my execution time by few 100 ms.
Can the outputs go wrong if I do not use clWaitForEvents? I would like to understand what might possibly go wrong if I do not use clWaitForEvents or clFinish.
Any pointers are appreciated.
Hopefully a slightly less complicated answer:
I've to launch 10 different kernels sequentially. Every kernel is dependent on the output of the previous kernel.
If you don't explicitly set CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE property in clCreateCommandQueue() call (= the usual case), it will be an in-order queue. You don't need to synchronize commands in them (actually you shouldn't, as you see it can considerably slow down execution). See the docs:
If the CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE property of a command-queue is not set, the commands enqueued to a command-queue execute in order. For example, if an application calls clEnqueueNDRangeKernel to execute kernel A followed by a clEnqueueNDRangeKernel to execute kernel B, the application can assume that kernel A finishes first and then kernel B is executed. If the memory objects output by kernel A are inputs to kernel B then kernel B will see the correct data in memory objects produced by execution of kernel A.
I would like to understand what might possibly go wrong if I do not use clWaitForEvents or clFinish.
If you're doing simple stuff on a single in-order queue, you don't need clWaitForEvents() at all. It's mostly useful if you want to wait for multiple events from multiple queues, or you're using out-of-order queues, or you want to enqueue 20 commands but wait for the 4th, or something similar.
For a single in-order queue, after clFinish() returns all commands will be completed and any&all events will have their status updated to complete or failed. So in the simplest case you don't need to deal with events at all, just enqueue everything you need (check the enqueues for errors though) and call clFinish().
Note that if you don't use any form of wait/flush (WaitForEvents / Finish / a blocking command), the implementation may take as much time as it wants to actually push those commands to a device. IOW you must either 1) use WaitForEvents or Finish, or 2) enqueue a blocking command (read/write/map/unmap) as the last command.
In-order-queue implicitly waits for each command completion in the order they are enqueued but only on device-side. This means host can't know what happened.
Out-of-order-queue does not guarantee any command order in anywhere and can have issues.
'Wait-for-event' waits on host side for an event of a command.
'Finish' waits on host side until all commands are complete.
'Non blocking buffer read/write' does not wait on host side.
'Blocking buffer read/write' waits on host side but does not wait for other commands.
Recommended solutions:
Inter-command sync (for using output of a command as input of next command)
in-order-queue.
or passing event of a command to another (if its an out-of-order queue)
Inter-queue(or out-of-order queue) sync (for overlapping buffer copies and kernel executions)
pass events from command(s) to another command
Device - host sync (for getting latest data to RAM(or getting first data from RAM) or pausing host)
enable blocking option on buffer commands
or add a clFinish
or use clWaitForEvent
Be informed when a command is complete(for reasons like benchmarking)
use event callback
or constantly query event state(CPU/pci-e usage increases)
Enqueueing 1 non-blocking buffer write + 1000 x kernels + 1 blocking buffer read on an in-order-queue can successfully execute a chain of 1000 kernels on initial data and get latest results on host side.
In a uniprocessor (UP) system, there's only one CPU core, so only one thread of execution can be happening at once. This thread of execution is synchronous (it gets a list of instructions in a queue and run them one by one). When we write code, it compiles to set of CPU instructions.
How can we have asynchronous behavior in software on a UP machine? Isn't everything just run in some fixed order chosen by the OS?
Even an out-of-order execution CPU gives the illusion of running instructions in program order. (This is separate from memory reordering observed by other cores or devices in the system. In a UP system, runtime memory reordering is only relevant for device drivers.)
An interrupt handler is a piece of code that runs asynchronously to the rest of the code, and can happen in response to an interrupt from a device outside the CPU. In user-space, a signal handler has equivalent semantics.
(Or a hardware interrupt can cause a context switch to another software thread. This is asynchronous as far as the software thread is concerned.)
Events like interrupts from network packets arriving or disk I/O completing happen asynchronously with respect to whatever the CPU was doing before the interrupt.
Asynchronous doesn't mean simultaneous, just that it can run between any two machine instructions of the rest of the code. A signal handler in a user-space program can run between any two machine instructions, so the code in the main program must work in a way that doesn't break if this happens.
e.g. A program with a signal-handler can't make any assumptions about data on the stack below the current stack pointer (i.e. in the un-reserved part of the stack). The red-zone in the x86-64 SysV ABI is a modification to this rule for user-space only, since the kernel can respect it when transferring control to a signal handler. The kernel itself can't use a red-zone, because hardware interrupts write to the stack outside of software control, before running the interrupt handler.
In an OS where I/O completion can result in the delivery of a POSIX signal (i.e. with POSIX async I/O), the timing of a signal can easily be determined by the timing of a hardware interrupts, so user-space code runs asynchronously with timing determined by things external to the computer. It's not just an issue for the kernel.
On a multicore system, there are obviously far more ways for things to happen in different orders more of the time.
Many processors are capable of multithreading, and many operating systems can simulate multithreading on single-threaded processors by swapping tasks in and out of the processor.
Lets say Unix is executing process A and an interrupt at higher level occurs. Then OS get a interrupt number and from IVT it looks up the routine to call.
Now how does the OS know that this interrupt was for process A and not for process B. It might have been that process B might have issued a disk read and it came back while OS was executing process A.
Thanks
Start with this: http://en.wikipedia.org/wiki/MINIX
Go buy the book and read it; it will really help a lot.
Interrupts aren't "for" processes. They're for devices and handled by device drivers.
The device driver handles the interrupt and updates the state of the device.
If the device driver concludes that an I/O operation is complete, it can then update the its queue of I/O requests to determine which operation completed. The operation is removed from the queue of pending operations.
The process which is waiting for that operation is now ready-to-run and can resume execution.
You are talking about a hardware interrupt and these are not targeted at processes.
If a process A requests a file, the filesystem layer, which already resides in the kernel, will fetch the file from the block device. The block device itself is handled by a driver.
When the interrupt occurs, triggered by the block device, the OS has this interrupt associated with the driver. So the driver is told to handle the interrupt. It will then query which blocks were read and see for what it requested them.
After the filesystem is told that the requested data is ready, it may further process it. Then, the process leaves blocked state.
In the next round of the scheduler, the scheduler may select to wake up this process. It may also select to wake up another process first.
As you can see, the interrupt occurance is fully disconnected from the process operation.
I always wondered what they are: every time I hear about them, images of futuristic flywheel-like devices go dancing (rolling?) through my mind...
What are they?
When you use regular locks (mutexes, critical sections etc), operating system puts your thread in the WAIT state and preempts it by scheduling other threads on the same core. This has a performance penalty if the wait time is really short, because your thread now has to wait for a preemption to receive CPU time again.
Besides, kernel objects are not available in every state of the kernel, such as in an interrupt handler or when paging is not available etc.
Spinlocks don't cause preemption but wait in a loop ("spin") till the other core releases the lock. This prevents the thread from losing its quantum and continue as soon as the lock gets released. The simple mechanism of spinlocks allows a kernel to utilize it in almost any state.
That's why on a single core machine a spinlock is simply a "disable interrupts" or "raise IRQL" which prevents thread scheduling completely.
Spinlocks ultimately allow kernels to avoid "Big Kernel Lock"s (a lock acquired when core enters kernel and released at the exit) and have granular locking over kernel primitives, causing better multi-processing on multi-core machines thus better performance.
EDIT: A question came up: "Does that mean I should use spinlocks wherever possible?" and I'll try to answer it:
As I mentioned, Spinlocks are only useful in places where anticipated waiting time is shorter than a quantum (read: milliseconds) and preemption doesn't make much sense (e.g. kernel objects aren't available).
If waiting time is unknown, or if you're in user mode Spinlocks aren't efficient. You consume 100% CPU time on the waiting core while checking if a spinlock is available. You prevent other threads from running on that core till your quantum expires. This scenario is only feasible for short bursts at kernel level and unlikely an option for a user-mode application.
Here is a question on SO addressing that: Spinlocks, How Useful Are They?
Say a resource is protected by a lock ,a thread that wants access to the resource needs to acquire the lock first. If the lock is not available, the thread might repeatedly check if the lock has been freed. During this time the thread busy waits, checking for the lock, using CPU, but not doing any useful work. Such a lock is termed as a spin lock.
It is pertty much a loop that keeps going till a certain condition is met:
while(cantGoOn) {};
while(something != TRUE ){};
// it happend
move_on();
It's a type of lock that does busy waiting
It's considered an anti-pattern, except for very low-level driver programming (where it can happen that calling a "proper" waiting function has more overhead than simply busy locking for a few cycles).
See for example Spinlocks in Linux kernel.
SpinLocks are the ones in which thread waits till the lock is available. This will normally be used to avoid overhead of obtaining the kernel objects when there is a scope of acquiring the kernel object within some small time period.
Ex:
While(SpinCount-- && Kernel Object is not free)
{}
try acquiring Kernel object
You would want to use a spinlock when you think it is cheaper to enter a busy waiting loop and pool a resource instead of blocking when the resource is locked.
Spinning can be beneficial when locks are fine grained and large in number (for example, a lock per node in a linked list) as well as when lock hold times are always extremely short. In general, while holding a spin lock, one should avoid blocking, calling anything that itself may block, holding more than one spin lock at once, making dynamically dispatched calls (interface and virtuals), making statically dispatched calls into any code one doesn't own, or allocating memory.
It's also important to note that SpinLock is a value type, for performance reasons. As such, one must be very careful not to accidentally copy a SpinLock instance, as the two instances (the original and the copy) would then be completely independent of one another, which would likely lead to erroneous behavior of the application. If a SpinLock instance must be passed around, it should be passed by reference rather than by value.
It's a loop that spins around until a condition is met.
In nutshell, spinlock employs atomic compare and swap (CAS) or test-and-set like instructions to implement lock free, wait free thread safe idiom. Such structures scale well in multi-core machines.
Well, yes - the point of spin locks (vs a traditional critical sections, etc) is that they offer better performance under some circumstances (multicore systems..), because they don't immediately yield the rest of the thread's quantum.
Spinlock, is a type of lock, which is non-block able & non-sleep-able. Any thread which want to acquire a spinlock for any shared or critical resource will continuously spin, wasting the CPU processing cycle till it acquire the lock for the specified resource. Once spinlock is acquired, it try to complete the work in its quantum and then release the resource respectively. Spinlock is the highest priority type of lock, simply can say, it is non-preemptive kind of lock.