It is said that one of the main benefits of Node (and presumable twisted et al) over more conventional threaded servers, is the very high concurrency enabled by the event loop model. The biggest reason for this is that each thread has a high memory footprint and swapping contexts is comparatively expensive. When you have thousands of threads the server spends most of its time swapping from thread to thread.
My question is, why don't operating systems or the underlying hardware support much more lightweight threads? If they did, could you solve the 10k problem with plain threads? If they can't, why is that?
Modern operating systems can support the execution of a very large number of threads.
More generally, hardware keeps getting faster (and recently, it has been getting faster in a way that is much friendlier to multithreading and multiprocessing than to single-threaded event loops - ie, increased number of cores, rather than increased processing throughput capabilities in a single core). If you can't afford the overhead of a thread today, you can probably afford it tomorrow.
What the cooperative multitasking systems of Twisted (and presumably Node.js et al) offers over pre-emptive multithreading (at least in the form of pthreads) is ease of programming.
Correctly using multithreading involves being much more careful than correctly using a single thread. An event loop is just the means of getting multiple things done without going beyond your single thread.
Considering the proliferation of parallel hardware, it would be ideal for multithreading or multiprocessing to get easier to do (and easier to do correctly). Actors, message passing, maybe even petri nets are some of the solutions people have attempted to solve this problem. They are still very marginal compared to the mainstream multithreading approach (pthreads). Another approach is SEDA, which uses multiple threads to run multiple event loops. This also hasn't caught on.
So, the people using event loops have probably decided that programmer time is worth more than CPU time, and the people using pthreads have probably decided the opposite, and the people exploring actors and such would like to value both kinds of time more highly (clearly insane, which is probably why no one listens to them).
The issue isn't really how heavyweight the threads are but the fact that to write correct multithreaded code you need locks on shared items and that prevents it from scaling with the number of threads because threads end up waiting for each other to gain locks and you rapidly reach the point where adding additional threads has no effect or even slows the system down as you get more lock contention.
In many cases you can avoid locking, but it's very difficult to get right, and sometimes you simply need a lock.
So if you are limited to a small number of threads, you might well find that removing the overhead of having to lock resources at all, or even think about it, makes a single threaded program faster than a multithreaded program no matter how many threads you add.
Basically locks can (depending on your program) be really expensive and can stop your program scaling beyond a few threads. And you almost always need to lock something.
It's not the overhead of a thread that's the problem, it's the synchronization between the threads. Even if you could switch between threads instantly, and had infinite memory none of that helps if each thread just ends up waiting in a queue for it's turn at some shared resource.
Related
What is overhead in term of parallel and concurrent programming (Haskell)?
However, even in a purely functional language, automatic parallelization is thwarted by an age-old problem: To make the program faster, we have to gain more from parallelism than we lose due to the overhead of adding it, and compile-time analysis cannot make good judgments in this area. An alternative approach is to use runtime profiling to find good candidates for parallelization and to feed this information back into the compiler. Even this, however, has not been terribly successful in practice.
(quoted from Simon Marlow's book Parallel and Concurrent Programming in Haskell)
What are some examples in Haskell?
In any system, a thread takes resources. You have to store the state of that thread somewhere. It takes time to create the thread and set it running. Now GHC uses lightweight "green threads", which are much less expensive than OS threads. But they still cost something.
If you were to (for example) spawn a new thread for every single add, subtract, multiply and divide... well, the work to spawn a new thread has to be at least several dozen machine instructions, whereas a trivial arithmetic operation is probably a single instruction. Queuing the work as sparks takes even less work than spawning a whole new thread, but even that isn't as cheap as just doing the operation on the current thread.
Basically the cost of the work you want to do in parallel has to exceed the cost of arranging to do it in parallel. (Whether that's launching an OS thread or a green thread or queuing a spark or whatever.) GHC has all sorts of stuff to lower the cost, but it's still not free.
You have to understand that threads are resources. They do not come for free. In other words: when you create a thread (independent of the language) then you have to make system calls, the OS has to create a thread instance, and so on. Threads have state - which changes over time; so some kind of thread management happens in the background.
And of course, when you end up with more threads than the underlying hardware can support - then the system will have to switch threads from time to time. Of course, that is not as expensive as switching full blown processes, but it still means that registers need to be saved (or restored), your hardware caches might be affected, and so on.
There are many solutions geared toward implementing "user-space" threads. Be it golang.org goroutines, python's green threads, C#'s async, erlang's processes etc. The idea is to allow concurrent programming even with a single or limited number of threads.
What I don't understand is, why are the OS threads so expensive? As I see it, either way you have to save the stack of the task (OS thread, or userland thread), which is a few tens of kilobytes, and you need a scheduler to move between two tasks.
The OS provides both of this functions for free. Why should OS threads be more expensive than "green" threads? What's the reason for the assumed performance degradation caused by having a dedicated OS thread for each "task"?
I want to amend Tudors answer which is a good starting point. There are two main overheads of threads:
Starting and stopping them. Involves creating a stack and kernel objects. Involves kernel transitions and global kernel locks.
Keeping their stack around.
(1) is only a problem if you are creating and stopping them all the time. This is solved commonly using thread pools. I consider this problem to be practically solved. Scheduling a task on a thread pool usually does not involve a trip to the kernel which makes it very fast. The overhead is on the order of a few interlocked memory operations and a few allocations.
(2) This becomes important only if you have many threads (> 100 or so). In this case async IO is a means to get rid of the threads. I found that if you don't have insane amounts of threads synchronous IO including blocking is slightly faster than async IO (you read that right: sync IO is faster).
Saving the stack is trivial, no matter what its size - the stack pointer needs to be saved in the Thread Info Block in the kernel, (so usualy saving most of the registers as well since they will have been pushed by whatever soft/hard interrupt caused the OS to be entered).
One issue is that a protection level ring-cycle is required to enter the kernel from user. This is an essential, but annoying, overhead. Then the driver or system call has to do whatever was requested by the interrupt and then the scheduling/dispatching of threads onto processors. If this results in the preemption of a thread from one process by a thread from another, a load of extra process context has to be swapped as well. Even more overhead is added if the OS decides that a thread that is running on another processor core than the one handling the interrupt mut be preempted - the other core must be hardware-interrupted, (this is on top of the hard/soft interrupt that entred the OS in the first place.
So, a scheduling run may be quite a complex operation.
'Green threads' or 'fibers' are, (usually), scheduled from user code. A context-change is much easier and cheaper than an OS interrupt etc. because no Wagnerian ring-cycle is required on every context-change, process-context does not change and the OS thread running the green thread group does not change.
Since something-for-nothing does not exist, there are problems with green threads. They ar run by 'real' OS threads. This means that if one 'green' thread in a group run by one OS thread makes an OS call that blocks, all green threads in the group are blocked. This means that simple calls like sleep() have to be 'emulated' by a state-machine that yields to other green threads, (yes, just like re-implementing the OS). Similarly, any inter-thread signalling.
Also, of course, green threads cannot directly respond to IO signaling, so somewhat defeating the point of having any threads in the first place.
There are many solutions geared toward implementing "user-space" threads. Be it golang.org goroutines, python's green threads, C#'s async, erlang's processes etc. The idea is to allow concurrent programming even with a single or limited number of threads.
It's an abstraction layer. It's easier for many people to grasp this concept and use it more effectively in many scenarios. It's also easier for many machines (assuming a good abstraction), since the model moves from width to pull in many cases. With pthreads (as an example), you have all the control. With other threading models, the idea is to reuse threads, for the process of creating a concurrent task to be inexpensive, and to use a completely different threading model. It's far easier to digest this model; there's less to learn and measure, and the results are generally good.
What I don't understand is, why are the OS threads so expensive? As I see it, either way you have to save the stack of the task (OS thread, or userland thread), which is a few tens of kilobytes, and you need a scheduler to move between two tasks.
Creating a thread is expensive, and the stack requires memory. As well, if your process is using many threads, then context switching can kill performance. So lightweight threading models became useful for a number of reasons. Creating an OS thread became a good solution for medium to large tasks, ideally in low numbers. That's restrictive, and quite time consuming to maintain.
A task/thread pool/userland thread does not need to worry about much of the context switching or thread creation. It's often "reuse the resource when it becomes available, if it's not ready now -- also, determine the number of active threads for this machine".
More commmonly (IMO), OS level threads are expensive because they are not used correctly by the engineers - either there are too many and there is a ton of context switching, there is competition for the same set of resources, the tasks are too small. It takes much more time to understand how to use OS threads correctly, and how to apply that best to the context of a program's execution.
The OS provides both of this functions for free.
They're available, but they are not free. They are complex, and very important to good performance. When you create an OS thread, it's given time 'soon' -- all the process' time is divided among the threads. That's not the common case with user threads. The task is often enqueued when the resource is not available. This reduces context switching, memory, and the total number of threads which must be created. When the task exits, the thread is given another.
Consider this analogy of time distribution:
Assume you are at a casino. There are a number people who want cards.
You have a fixed number of dealers. There are fewer dealers than people who want cards.
There is not always enough cards for every person at any given time.
People need all cards to complete their game/hand. They return their cards to the dealer when their game/hand is complete.
How would you ask the dealers to distribute cards?
Under the OS scheduler, that would be based on (thread) priority. Every person would be given one card at a time (CPU time), and priority would be evaluated continually.
The people represent the task or thread's work. The cards represent time and resources. The dealers represent threads and resources.
How would you deal fastest if there were 2 dealers and 3 people? and if there were 5 dealers and 500 people? How could you minimize running out of cards to deal? With threads, adding cards and adding dealers is not a solution you can deliver 'on demand'. Adding CPUs is equivalent to adding dealers. Adding threads is equivalent to dealers dealing cards to more people at a time (increases context switching). There are a number of strategies to deal cards more quickly, especially after you eliminate the people's need for cards in a certain amount of time. Would it not be faster to go to a table and deal to a person or people until their game is complete if the dealer to people ratio were 1/50? Compare this to visiting every table based on priority, and coordinating visitation among all dealers (the OS approach). That's not to imply the OS is stupid -- it implies that creating an OS thread is an engineer adding more people and more tables, potentially more than the dealers can reasonably handle. Fortunately, the constraints may be lifted in many cases by using other multithreading models and higher abstractions.
Why should OS threads be more expensive than "green" threads? What's the reason for the assumed performance degradation caused by having a dedicated OS thread for each "task"?
If you developed a performance critical low level threading library (e.g. upon pthreads), you would recognize the importance of reuse (and implement it in your library as a model available for users). From that angle, the importance of higher level multithreading models is a simple and obvious solution/optimization based on real world usage as well as the ideal that the entry bar for adopting and effectively utilizing multithreading can be lowered.
It's not that they are expensive -- the lightweight threads' model and pool is a better solution for many problems, and a more appropriate abstraction for engineers who do not understand threads well. The complexity of multithreading is greatly simplified (and often more performant in real world usage) under this model. With OS threads, you do have more control, but several more considerations must be made to use them as effectively as possible -- heeding these consideration can dramatically reflow a program's execution/implementation. With higher level abstractions, many of these complexities are minimized by completely altering the flow of task execution (width vs pull).
The problem with starting kernel threads for each small task is that it incurs a non-negligible overhead to start and stop, coupled with the stack size it needs.
This is the first important point: thread pools exist so that you can recycle threads, in order to avoid wasting time starting them as well as wasting memory for their stacks.
Secondly, if you fire off threads to do asynchronous I/O, they will spend most of their time blocked waiting for the I/O to complete, thus effectively not doing any work and wasting memory. A much better option is to have a single worker handle multiple async calls (through some under-the-hood scheduling technique, such as multiplexing), thus again saving memory and time.
One thing that makes "green" threads faster than kernel threads is that they are user-space objects, managed by a virtual machine. Starting them is a user space call, while starting a thread is a kernel-space call that is much slower.
A person in Google shows an interesting approach.
According to him, kernel mode switching itself is not the bottleneck, and the core cost happen on SMP scheduler. And he claims M:N schedule assisted by kernel wouldn't be expensive, and this makes me to expect general M:N threading to be available on every languages.
Because the OS. Imagine that instead of asking you to clean the house your grandmother has to call the social service that does some paperwork and a week after assigns a social worker for helping her. The worker can be called off at any time and replaced with another one, which again takes several days.
That's pretty ineffective and slow, huh?
In this metaphor you are a userland coroutine scheduler, the social service is an OS with its kernel-level thread scheduler, and a social worker is a fully-fledged thread.
I think the two things are in different levels.
Thread or Process is an instance of the program which is being executed. In a process/thread there is much more things in it. Execution stack, opening files, signals, processors status, and a many other things.
Greentlet is different, it is runs in vm. It supplies a light-weight thread. Many of them supply a pseudo-concurrently (typically in a single or a few OS-level threads). And often they supply a lock-free method by data-transmission instead of data sharing.
So, the two things focus different, so the weight are different.
And In my mind, the greenlet should be finished in the VM not the OS.
Do you think threadpooling design pattern is the way to go for the multi-core future?
A threadpooling library for instance, if used extensively, makes/force the application writer
(1) to break the problem into separate parallel jobs hence promoting (enforcing :) ) parallelism
(2) With abstraction from all the low level OS calls, synchronization etc etc makes programmer's life easier. (Especially for C programmers :) )
I have strong belief its the best way (Or One of the "best" ways :) ) for multi-core future...
So, my question is, am I write in thinking so or I am in some delusion :)
Regards,
Microkernel
Thread pooling is a technique that involves a queue and a number of threads taking jobs from the queue and process them. This is in contrast to the technique of just starting new threads whenever a new task arrives.
Benefits are that the maximum number of threads is limited to avoid too much threading and that there is less overhead involved with any new task (Thread is already running and takes task. No threat starting needed).
Whether this is a good design highly depends on your problem. If you have many short jobs that come to your program at a very fast rate, then this is a good idea because the lower overhead is really a benefit. If you have extremely large numbers of concurrent tasks this is a good idea to keep your scheduler from having to do too much work.
There are many areas where thread pooling is just not helpful. So you cannot generalize. Sometimes multi threading at all is not possible. Or not even desired, as multi threading adds an unpredictable element (race conditions) to your code which is extremely hard to debug.
A thread pooling library can hardly "force" you to use it. You still need to think stuff through and if you just start one thread... Won't help.
As almost every informatics topic the answer is: It Depends.
the pooling system is fine with Embarrassingly parallel http://en.wikipedia.org/wiki/Embarrassingly_parallel
For other task where you need more syncornization between threads it's not that good
For the Windows NT engine thread pools are usually much less efficient than I/O Completion Ports. These are covered extensively in numerous questions and answers here. IOCPs enable event-driven processing in that multiple threads can wait on the IOCP until an event occurs due to an IOC (read or write) on a socket or handle which is then queued to the IOCP. The IOCP in turn pairs a waiting thread with the id of the event and releases the thread for processing. After the thread has processed the event and initiated a new I/O it returns to the IOCP to wait for the next event (which may or may not be the completion of the I/O it just initiated).
IOCs may also be artificially signalled by explicit posting from a non-event.
Using IOCPs is not polling. The optimal IOCP implementation will have as many threads waiting on the IOCP as there are cores in the system. The threads may all execute the same physical code if that is deemed efficient. Since a thread processes from an IOC up until it issues an I/O it does nothing which forces it to wait for other resources except perhaps to compete for access to thread-safe areas. It is a natural choice to move away from the "one handle per thread" paradigm. IOCP-controlled threads are therefore as efficient as the programmer is able to construct them.
I like the answer by #yaankee a lot except I would argue that thread pool is almost always the right way to go. The reason: a thread pool can degenerate itself into a simple static work partitioning model for problems like matrix-matrix multiply. OpenMP guided is kind of along those lines.
As far as I'm concerned, the ideal amount of threads is 3: one for the UI, one for CPU resources, and one for IO resources.
But I'm probably wrong.
I'm just getting introduced to them, but I've always used one for the UI and one for everything else.
When should I use threads and how? How do I know if I should be using them?
Unfortunately, there are no hard and fast rules to using Threads. If you have too many threads the processor will spend all its time generating and switching between them. Use too few threads you will not get the throughput you want in your application. Additionally using threads is not easy. A language like C# makes it easier on you because you have tools like ThreadPool.QueueUserWorkItem. This allows the system to manage thread creation and destruction. This helps mitigate the overhead of creating a new thread to pass the work onto. You have to remember that the creation of a thread is not an operation that you get for "free." There are costs associated with starting a thread so that should always be taken into consideration.
Depending upon the language you are using to write your application you will dictate how much you need to worry about using threads.
The times I find most often that I need to consider creating threads explicitly are:
Asynchronous operations
Operations that can be parallelized
Continual running background operations
The answer totally depends on what you're planning on doing. However, one for CPU resources is a bad move - your CPU may have up to six cores, plus hyperthreading, in a retail CPU, and most CPUs will have two or more. In this case, you should have as many threads as CPU cores, plus a few more for scheduling mishaps. The whole CPU is not a single-threaded beast, it may have many cores and need many threads for 100% utilization.
You should use threads if and only if your target demographic will virtually all have multi-core (as is the case in current desktop/laptop markets), and you have determined that one core is not enough performance.
Herb Sutter wrote an article for Dr. Dobb's Journal in which he talks about the three pillars of concurrency. This article does a very good job of breaking down which problems are good candidates for being solved via threading constructs.
From the SQLite FAQ: "Threads are evil. Avoid Them." Only use them when you absolutely have to.
If you have to, then take steps to avoid the usual carnage. Use thread pools to execute fine-grained tasks with no interdependencies, using GUI-framework-provided facilities to dispatch outcomes back to the UI. Avoid sharing data between long-running threads; use message queues to pass information between them (and to synchronise).
A more exotic solution is to use languages such as Erlang that are explicit designed for fine-grained parallelism without sacrificing safety and comprehensibility. Concurrency itself is of fundamental importance to the future of computation; threads are simply a horrible, broken way to express it.
The "ideal number of threads" depends on your particular problem and how much parallelism you can exploit. If you have a problem that is "embarassingly parallel" in that it can be subdivided into independent problems with little to no communication between them required, and you have enough cores that you can actually get true parallelism, then how many threads you use depends on things like the problem size, the cache line size, the context switching and spawning overhead, and various other things that is really hard to compute before hand. For such situations, you really have to do some profiling in order to choose an optimal sharding/partitioning of your problem across threads. It typically doesn't make sense, though, to use more threads than you do cores. It is also true that if you have lots of synchronization, then you may, in fact, have a performance penalty for using threads. It's highly dependent on the particular problem as well as how interdependent the various steps are. As a guiding principle, you need to be aware that spawning threads and thread synchronization are expensive operations, but performing computations in parallel can increase throughput if communication and other forms of synchronization is minimal. You should also be aware that threading can lead to very poor cache performance if your threads end up invalidating a mutually shared cache line.
I've been toying around with the Parallel library in .NET 4.0. Recently, I developed a custom ORM for some unusual read/write operations one of our large systems has to use. This allows me to decorate an object with attributes and have reflection figure out what columns it has to pull from the database, as well as what XML it has to output on writes.
Since I envision this wrapper to be reused in many projects, I'd like to squeeze as much speed out of it as possible. This library will mostly be used in .NET web applications. I'm testing the framework using a throwaway console application to poke at the classes I've created.
I've now learned a lesson of the overhead that multithreading comes with. Multithreading causes it to run slower. From reading around, it seems like it's intuitive to people who've been doing it for a long time, but it's actually counter-intuitive to me: how can running a method 30 times at the same time be slower than running it 30 times sequentially?
I don't think I'm causing problems by multiple threads having to fight over the same shared object (though I'm not good enough at it yet to tell for sure or not), so I assume the slowdown is coming from the overhead of spawning all those threads and the runtime keeping them all straight. So:
Though I'm doing it mainly as a learning exercise, is this pessimization? For trivial, non-IO tasks, is multithreading overkill? My main goal is speed, not responsiveness of the UI or anything.
Would running the same multithreading code in IIS cause it to speed up because of already-created threads in the thread pool, whereas right now I'm using a console app, which I assume would be single-threaded until I told it otherwise? I'm about to run some tests, but I figure there's some base knowledge I'm missing to know why it would be one way or the other. My console app is also running on my desktop with two cores, whereas a server for a web app would have more, so I might have to use that as a variable as well.
Thread's don't actually all run concurrently.
On a desktop machine I'm presuming you have a dual core CPU, (maybe a quad at most). This means only 2/4 threads can be running at the same time.
If you have spawned 30 threads, the OS is going to have to context switch between those 30 threads to keep them all running. Context switches are quite costly, so hence the slowdown.
As a basic suggestion, I'd aim for 1 thread per CPU if you are trying to optimise calculations. Any more than this and you're not really doing any extra work, you are just swapping threads in an out on the same CPU. Try to think of your computer as having a limited number of workers inside, you can't do more work concurrently than the number of workers you have available.
Some of the new features in the .net 4.0 parallel task library allow you to do things that account for scalability in the number of threads. For example you can create a bunch of tasks and the task parallel library will internally figure out how many CPUs you have available, and optimise the number of threads is creates/uses so as not to overload the CPUs, so you could create 30 tasks, but on a dual core machine the TP library would still only create 2 threads, and queue the . Obviously, this will scale very nicely when you get to run it on a bigger machine. Or you can use something like ThreadPool.QueueUserWorkItem(...) to queue up a bunch of tasks, and the pool will automatically manage how many threads is uses to perform those tasks.
Yes there is a lot of overhead to thread creation, but if you are using the .net thread pool, (or the parallel task library in 4.0) .net will be managing your thread creation, and you may actually find it creates less threads than the number of tasks you have created. It will internally swap your tasks around on the available threads. If you actually want to control explicit creation of actual threads you would need to use the Thread class.
[Some cpu's can do clever stuff with threads and can have multiple Threads running per CPU - see hyperthreading - but check out your task manager, I'd be very surprised if you have more than 4-8 virtual CPUs on today's desktops]
There are so many issues with this that it pays to understand what is happening under the covers. I would highly recommend the "Concurrent Programming on Windows" book by Joe Duffy and the "Java Concurrency in Practice" book. The latter talks about processor architecture at the level you need to understand it when writing multithreaded code. One issue you are going to hit that's going to hurt your code is caching, or more likely the lack of it.
As has been stated there is an overhead to scheduling and running threads, but you may find that there is a larger overhead when you share data across threads. That data may be flushed from the processor cache into main memory, and that will cause serious slow downs to your code.
This is the sort of low-level stuff that managed environments are supposed to protect us from, however, when writing highly parallel code, this is exactly the sort of issue you have to deal with.
A colleague of mine recorded a screencast about the performance issue with Parallel.For and Parallel.ForEach which may help:
http://rocksolidknowledge.com/ScreenCasts.mvc/Watch?video=ParallelLoops.wmv
You're speaking of an ORM, so I presume some amount of I/O is going on. If this is the case, the overhead of thread creation and context switching is going to be comparatively non-existent.
Most likely, you're experiencing I/O contention: it can be slower (particularly on rotational hard drives, but also on other storage devices) to read the same set of data if you read it out of order than if you read it in-order. So, if you're executing 30 database queries, it's possible they'll run faster sequentially than in parallel if they're all backed by the same I/O device and the queries aren't in cache. Running them in parallel may cause the system to have a bunch of I/O read requests almost simultaneously, which may cause the OS to read little bits of each in turn - causing your drive head to jump back and forth, wasting precious milliseconds.
But that's just a guess; it's not possible to really determine what's causing your slowdown without knowing more.
Although thread creation is "extremely expensive" when compared to say adding two numbers, it's not usually something you'll easily overdo. If your operations are extremely short (say, a millisecond or less), using a thread-pool rather than new threads will noticeably save time. Generally though, if your operations are that short, you should reconsider the granularity of parallelism anyhow; perhaps you're better off splitting the computation into bigger chunks: for instance, by having a fairly low number of worker tasks which handle entire batches of smaller work-items at a time rather than each item separately.