Advice on starting a large multi-threaded programming project - multithreading

My company currently runs a third-party simulation program (natural catastrophe risk modeling) that sucks up gigabytes of data off a disk and then crunches for several days to produce results. I will soon be asked to rewrite this as a multi-threaded app so that it runs in hours instead of days. I expect to have about 6 months to complete the conversion and will be working solo.
We have a 24-proc box to run this. I will have access to the source of the original program (written in C++ I think), but at this point I know very little about how it's designed.
I need advice on how to tackle this. I'm an experienced programmer (~ 30 years, currently working in C# 3.5) but have no multi-processor/multi-threaded experience. I'm willing and eager to learn a new language if appropriate. I'm looking for recommendations on languages, learning resources, books, architectural guidelines. etc.
Requirements: Windows OS. A commercial grade compiler with lots of support and good learning resources available. There is no need for a fancy GUI - it will probably run from a config file and put results into a SQL Server database.
Edit: The current app is C++ but I will almost certainly not be using that language for the re-write. I removed the C++ tag that someone added.

Numerical process simulations are typically run over a single discretised problem grid (for example, the surface of the Earth or clouds of gas and dust), which usually rules out simple task farming or concurrency approaches. This is because a grid divided over a set of processors representing an area of physical space is not a set of independent tasks. The grid cells at the edge of each subgrid need to be updated based on the values of grid cells stored on other processors, which are adjacent in logical space.
In high-performance computing, simulations are typically parallelised using either MPI or OpenMP. MPI is a message passing library with bindings for many languages, including C, C++, Fortran, Python, and C#. OpenMP is an API for shared-memory multiprocessing. In general, MPI is more difficult to code than OpenMP, and is much more invasive, but is also much more flexible. OpenMP requires a memory area shared between processors, so is not suited to many architectures. Hybrid schemes are also possible.
This type of programming has its own special challenges. As well as race conditions, deadlocks, livelocks, and all the other joys of concurrent programming, you need to consider the topology of your processor grid - how you choose to split your logical grid across your physical processors. This is important because your parallel speedup is a function of the amount of communication between your processors, which itself is a function of the total edge length of your decomposed grid. As you add more processors, this surface area increases, increasing the amount of communication overhead. Increasing the granularity will eventually become prohibitive.
The other important consideration is the proportion of the code which can be parallelised. Amdahl's law then dictates the maximum theoretically attainable speedup. You should be able to estimate this before you start writing any code.
Both of these facts will conspire to limit the maximum number of processors you can run on. The sweet spot may be considerably lower than you think.
I recommend the book High Performance Computing, if you can get hold of it. In particular, the chapter on performance benchmarking and tuning is priceless.
An excellent online overview of parallel computing, which covers the major issues, is this introduction from Lawerence Livermore National Laboratory.

Your biggest problem in a multithreaded project is that too much state is visible across threads - it is too easy to write code that reads / mutates data in an unsafe manner, especially in a multiprocessor environment where issues such as cache coherency, weakly consistent memory etc might come into play.
Debugging race conditions is distinctly unpleasant.
Approach your design as you would if, say, you were considering distributing your work across multiple machines on a network: that is, identify what tasks can happen in parallel, what the inputs to each task are, what the outputs of each task are, and what tasks must complete before a given task can begin. The point of the exercise is to ensure that each place where data becomes visible to another thread, and each place where a new thread is spawned, are carefully considered.
Once such an initial design is complete, there will be a clear division of ownership of data, and clear points at which ownership is taken / transferred; and so you will be in a very good position to take advantage of the possibilities that multithreading offers you - cheaply shared data, cheap synchronisation, lockless shared data structures - safely.

If you can split the workload up into non-dependent chunks of work (i.e., the data set can be processed in bits, there aren't lots of data dependencies), then I'd use a thread pool / task mechanism. Presumably whatever C# has as an equivalent to Java's java.util.concurrent. I'd create work units from the data, and wrap them in a task, and then throw the tasks at the thread pool.
Of course performance might be a necessity here. If you can keep the original processing code kernel as-is, then you can call it from within your C# application.
If the code has lots of data dependencies, it may be a lot harder to break up into threaded tasks, but you might be able to break it up into a pipeline of actions. This means thread 1 passes data to thread 2, which passes data to threads 3 through 8, which pass data onto thread 9, etc.
If the code has a lot of floating point mathematics, it might be worth looking at rewriting in OpenCL or CUDA, and running it on GPUs instead of CPUs.

For a 6 month project I'd say it definitely pays out to start reading a good book about the subject first. I would suggest Joe Duffy's Concurrent Programming on Windows. It's the most thorough book I know about the subject and it covers both .NET and native Win32 threading. I've written multithreaded programs for 10 years when I discovered this gem and still found things I didn't know in almost every chapter.
Also, "natural catastrophe risk modeling" sounds like a lot of math. Maybe you should have a look at Intel's IPP library: it provides primitives for many common low-level math and signal processing algorithms. It supports multi threading out of the box, which may make your task significantly easier.

There are a lot of techniques that can be used to deal with multithreading if you design the project for it.
The most general and universal is simply "avoid shared state". Whenever possible, copy resources between threads, rather than making them access the same shared copy.
If you're writing the low-level synchronization code yourself, you have to remember to make absolutely no assumptions. Both the compiler and CPU may reorder your code, creating race conditions or deadlocks where none would seem possible when reading the code. The only way to prevent this is with memory barriers. And remember that even the simplest operation may be subject to threading issues. Something as simple as ++i is typically not atomic, and if multiple threads access i, you'll get unpredictable results.
And of course, just because you've assigned a value to a variable, that's no guarantee that the new value will be visible to other threads. The compiler may defer actually writing it out to memory. Again, a memory barrier forces it to "flush" all pending memory I/O.
If I were you, I'd go with a higher level synchronization model than simple locks/mutexes/monitors/critical sections if possible. There are a few CSP libraries available for most languages and platforms, including .NET languages and native C++.
This usually makes race conditions and deadlocks trivial to detect and fix, and allows a ridiculous level of scalability. But there's a certain amount of overhead associated with this paradigm as well, so each thread might get less work done than it would with other techniques. It also requires the entire application to be structured specifically for this paradigm (so it's tricky to retrofit onto existing code, but since you're starting from scratch, it's less of an issue -- but it'll still be unfamiliar to you)
Another approach might be Transactional Memory. This is easier to fit into a traditional program structure, but also has some limitations, and I don't know of many production-quality libraries for it (STM.NET was recently released, and may be worth checking out. Intel has a C++ compiler with STM extensions built into the language as well)
But whichever approach you use, you'll have to think carefully about how to split the work up into independent tasks, and how to avoid cross-talk between threads. Any time two threads access the same variable, you have a potential bug. And any time two threads access the same variable or just another variable near the same address (for example, the next or previous element in an array), data will have to be exchanged between cores, forcing it to be flushed from CPU cache to memory, and then read into the other core's cache. Which can be a major performance hit.
Oh, and if you do write the application in C++, don't underestimate the language. You'll have to learn the language in detail before you'll be able to write robust code, much less robust threaded code.

One thing we've done in this situation that has worked really well for us is to break the work to be done into individual chunks and the actions on each chunk into different processors. Then we have chains of processors and data chunks can work through the chains independently. Each set of processors within the chain can run on multiple threads each and can process more or less data depending on their own performance relative to the other processors in the chain.
Also breaking up both the data and actions into smaller pieces makes the app much more maintainable and testable.

There's plenty of specific bits of individual advice that could be given here, and several people have done so already.
However nobody can tell you exactly how to make this all work for your specific requirements (which you don't even fully know yourself yet), so I'd strongly recommend you read up on HPC (High Performance Computing) for now to get the over-arching concepts clear and have a better idea which direction suits your needs the most.

The model you choose to use will be dictated by the structure of your data. Is your data tightly coupled or loosely coupled? If your simulation data is tightly coupled then you'll want to look at OpenMP or MPI (parallel computing). If your data is loosely coupled then a job pool is probably a better fit... possibly even a distributed computing approach could work.
My advice is get and read an introductory text to get familiar with the various models of concurrency/parallelism. Then look at your application's needs and decide which architecture you're going to need to use. After you know which architecture you need, then you can look at tools to assist you.
A fairly highly rated book which works as an introduction to the topic is "The Art of Concurrency: A Thread Monkey's Guide to Writing Parallel Application".

Read about Erlang and the "Actor Model" in particular. If you make all your data immutable, you will have a much easier time parallelizing it.

Most of the other answers offer good advice regarding partitioning the project - look for tasks that can be cleanly executed in parallel with very little data sharing required. Be aware of non-thread safe constructs such as static or global variables, or libraries that are not thread safe. The worst one we've encountered is the TNT library, which doesn't even allow thread-safe reads under some circumstances.
As with all optimisation, concentrate on the bottlenecks first, because threading adds a lot of complexity you want to avoid it where it isn't necessary.
You'll need a good grasp of the various threading primitives (mutexes, semaphores, critical sections, conditions, etc.) and the situations in which they are useful.
One thing I would add, if you're intending to stay with C++, is that we have had a lot of success using the boost.thread library. It supplies most of the required multi-threading primitives, although does lack a thread pool (and I would be wary of the unofficial "boost" thread pool one can locate via google, because it suffers from a number of deadlock issues).

I would consider doing this in .NET 4.0 since it has a lot of new support specifically targeted at making writing concurrent code easier. Its official release date is March 22, 2010, but it will probably RTM before then and you can start with the reasonably stable Beta 2 now.
You can either use C# that you're more familiar with or you can use managed C++.
At a high level, try to break up the program into System.Threading.Tasks.Task's which are individual units of work. In addition, I'd minimize use of shared state and consider using Parallel.For (or ForEach) and/or PLINQ where possible.
If you do this, a lot of the heavy lifting will be done for you in a very efficient way. It's the direction that Microsoft is going to increasingly support.
2: I would consider doing this in .NET 4.0 since it has a lot of new support specifically targeted at making writing concurrent code easier. Its official release date is March 22, 2010, but it will probably RTM before then and you can start with the reasonably stable Beta 2 now. At a high level, try to break up the program into System.Threading.Tasks.Task's which are individual units of work. In addition, I'd minimize use of shared state and consider using Parallel.For and/or PLINQ where possible. If you do this, a lot of the heavy lifting will be done for you in a very efficient way. 1: http://msdn.microsoft.com/en-us/library/dd321424%28VS.100%29.aspx

Sorry i just want to add a pessimistic or better realistic answer here.
You are under time pressure. 6 month deadline and you don't even know for sure what language is this system and what it does and how it is organized. If it is not a trivial calculation then it is a very bad start.
Most importantly: You say you have never done mulitithreading programming before. This is where i get 4 alarm clocks ringing at once. Multithreading is difficult and takes a long time to learn it when you want to do it right - and you need to do it right when you want to win a huge speed increase. Debugging is extremely nasty even with good tools like Total Views debugger or Intels VTune.
Then you say you want to rewrite the app in another lanugage - well this isn't as bad as you have to rewrite it anyway. THe chance to turn a single threaded Program into a well working multithreaded one without total redesign is almost zero.
But learning multithreading and a new language (what is your C++ skills?) with a timeline of 3 month (you have to write a throw away prototype - so i cut the timespan into two halfs) is extremely challenging.
My advise here is simple and will not like it: Learn multithreadings now - because it is a required skill set in the future - but leave this job to someone who already has experience. Well unless you don't care about the program being successfull and are just looking for 6 month payment.

If it's possible to have all the threads working on disjoint sets of process data, and have other information stored in the SQL database, you can quite easily do it in C++, and just spawn off new threads to work on their own parts using the Windows API. The SQL server will handle all the hard synchronization magic with its DB transactions! And of course C++ will perform a lot faster than C#.
You should definitely revise C++ for this task, and understand the C++ code, and look for efficiency bugs in the existing code as well as adding the multi-threaded functionality.

You've tagged this question as C++ but mentioned that you're a C# developer currently, so I'm not sure if you'll be tackling this assignment from C++ or C#. Anyway, in case you're going to be using C# or .NET (including C++/CLI): I have the following MSDN article bookmarked and would highly recommend reading through it as part of your prep work.
Calling Synchronous Methods Asynchronously

Whatever technology your going to write this, take a look a this must read book on concurrency "Concurrent programming in Java" and for .Net I highly recommend the retlang library for concurrent app.

I don't know if it was mentioned yet, but if I were in your shoes, what I would be doing right now (aside from reading every answer posted here) is writing a multiple threaded example application in your favorite (most used) language.
I don't have extensive multithreaded experience. I've played around with it in the past for fun but I think gaining some experience with a throw-away application will suit your future efforts.
I wish you luck in this endeavor and I must admit I wish I had the opportunity to work on something like this...

Related

Clojure: Create and manage multiple threads

I wrote a program which needs to process a very large dataset and I'm planning to run it with multiple threads in a high-end machine.
I'm a beginner in Clojure and i'm lost in the myriad of tools at disposal -
agents, futures, core.async (and Quartzite?). I would like to know which one is most suited for this job.
The following describes my situation:
I have a function which transforms some data and store it in database.
The argument to the said function is popped from a Redis set.
Run the function in several separate threads as long as there is a value in the Redis set.
For simplicity, futures can't be beat. They create a new thread, and return a value from it. However, often you need more fine-grained control than they provide.
The core.async library has nice support for parallelism (via pipeline, see below), and it also provides automatic back-pressure. You have to have a way to control the flow of data such that no one's starving for work, or burdened by too much of it. core.async channels must be bounded, and this helps with this problem. Also, it's a pretty logical model of your problem: taking a value from a source, transforming it (maybe using a transducer?) with some given parallelism, and then putting the result to your database.
You can also go the manual route of using Java's excellent j.u.concurrent library. There are low level primitives as well as thread management tools for thread pools. All of this is accessible within clojure.
From a design standpoint, it comes down to whether you are more CPU-bound or I/O-bound. This affects decisions such as whether or not you will perform parallel reads from redis and writes to your database. If you are CPU-bound and thus your bottleneck is the computation, then it wouldn't make much sense to parallelize your reads from redis, or your writes to your database, would it? These are the types of things to consider.
You really have two problems to solve: (1) your familiarity with clojure's/java's concurrency mechanisms, and (2) your approach to this problem (i.e., how would you approach this problem, irrespective of the language you're using?). Once you solve #2, you will have a much better idea of which tools to use that I mentioned above, and how to use them.
Sounds like you may have a
good
embarrassingly parallel problem
to solve. In that case, you could start simply by coding up your
processing into a top-level function that processes the first datum.
Once that's working, wrap it in
a map to handle all of the
data sequentially (serially, one-at-a-time).
You might want to start tackling the bigger problem with just a few
items from your data set. That will make your testing smoother and
faster.
After you have the map working, it's time to just add a p
(parallel) to your code to make it
a pmap. This is a very
rewarding way to heat up your
machine.
Here is
a discussion about the number of threads pmap uses.
The above is the simplest approach. If you need finer control over
the concurrency, the
this concurrency screencast explores
the use cases.
It is hard to be precise w/o knowing the details of your problem. There are several choices as you mention:
Plain Java threads & threadpools. If your problem is similar to a pre-existing Java solution, this may be the most straightforward.
Simple Clojure threading with future et al. Kicking off a thread with future and getting the result in a promise is very easy.
Replace map with pmap (parallel map). This can help in simple cases that are primarily map/reduce oriented.
The Claypoole library: Lots of tools to make multithreading simpler and easier. Please see their GitHub project and the Clojure/West talk.

Are there any practical alternatives to threads?

While reading up on SQLite, I stumbled upon this quote in the FAQ: "Threads are evil. Avoid them."
I have a lot of respect for SQLite, so I couldn't just disregard this. I got thinking what else I could, according to the "avoid them" policy, use instead in order to parallelize my tasks. As an example, the application I'm currently working on requires a user interface that is always responsive, and needs to poll several websites from time to time (a process which takes at least 30 seconds for each website).
So I opened up the PDF linked from that FAQ, and essentially it seems that the paper suggests several techniques to be applied together with threads, such as barriers or transactional memory - rather than any techniques to replace threads altogether.
Given that these techniques do not fully dispense with threads (unless I misunderstood what the paper is saying), I can see two options: either the SQLite FAQ does not literally mean what it says, or there exist practical approaches that actually avoid the use of threads altogether. Are there any?
Just a quick note on tasklets/cooperative scheduling as an alternative - this looks great in small examples, but I wonder whether a large-ish UI-heavy application can be practically parallelized in a solely cooperative way. If you have done this successfully or know of such examples this certainly qualifies as a valid answer!
Note: This answer no longer accurately reflects what I think about this subject. I don't like its overly dramatic, somewhat nasty tone. Also, I am not so certain that the quest for provably correct software has been so useless as I seemed to think back then. I am leaving this answer up because it is accepted, and up-voted, and to edit it into something I currently believe would pretty much vandalize it.
I finally got around to reading the paper. Where do I start?
The author is singing an old song, which goes something like this: "If you can't prove the program is correct, we're all doomed!" It sounds best when screamed loudly accompanied by over modulated electric guitars and a rapid drum beat. Academics started singing that song when computer science was in the domain of mathematics, a world where if you don't have a proof, you don't have anything. Even after the first computer science department was cleaved from the mathematics department, they kept singing that song. They are singing that song today, and nobody is listening. Why? Because the rest of us are busy creating useful things, good things out of software that can't be proved correct.
The presence of threads makes it even more difficult to prove a program correct, but who cares? Even without threads, only the most trivial of programs can be proved correct. Why do I care if my non-trivial program, which could not be proved correct, is even more unprovable after I use threading? I don't.
If you weren't sure the author was living in an academic dreamworld, you can be sure of it after he maintains that the coordination language he suggests as an alternative to threads could best be expressed with a "visual syntax" (drawing graphs on the screen). I've never heard that suggestion before, except every year of my career. A language that can only be manipulated by GUI and does not play with any of the programmer's usual tools is not an improvement. The author goes on to cite UML as a shining example of a visual syntax which is "routinely combined with C++ and Java." Routinely in what world?
In the mean time, I and many other programmers go on using threads without all that much trouble. How to use threads well and safely is pretty much a solved problem, as long as you don't get all hung up on provability.
Look. Threading is a big kid's toy, and you do need to know some theory and usage patterns to use them well. Just as with databases, distributed processing, or any of the other beyond-grade-school devices that programmers successfully use every day. But just because you can't prove it correct doesn't mean it's wrong.
The statement in the SQLite FAQ, as I read it, is just a comment on how difficult threading can be to the uninitiated. It is the author's opinion, and it might be a valid one. But saying you should never use threads is throwing the baby out with the bath water, in my opinion. Threads are a tool. Like all tools, they can be used and they can be abused. I can read his paper and be convinced that threads are the devil, but I have used them successfully, without killing kittens.
Keep in mind that SQLite is written to be as lightweight and easy to understand (from a coding standpoint) as possible, so I would imagine that threading is kind of the antithesis to this lightweight approach.
Also, SQLite is not meant to be used in a highly-concurrent environment. If you have one of these, you might be better off working with a more enterprisey database like Postgres.
Evil, but a necessary evil. High level abstractions of threads (Tasks in .NET for example) are becoming more common but for the most part the industry is not trying to find a way to avoid threads, just making it easier to deal with the complexities that come with any kind of concurrent programming.
One trend I've noticed, at least in the Cocoa domain, is help from the framework. Apple has gone to great lengths to help developers with the relatively difficult concept of concurrent programming. Some things I've seen:
Different granularity of threading. Cocoa supports everything from posix threads (low level) to object oriented threading with NSLock and NSThread, to high level parellelism such as NSOperation. Depending on your task, using a high level tool like NSOperation is easier and gets the job done.
Threading behind the scenes via an API. Lots of the UI and animation stuff in cocoa is hidden behind an API. You are responsible for calling an API method and providing an asynchronous callback this executed when the secondary thread completes (for example the end of some animation).
openMP. There are tools like openMP that allow you to provide pragmas that describe to the compiler that some task may be safely parelellized. For example iterating a set of items in an independent way.
It seems like a big push in this industry is to make things simple for the Application developers and leave the gory thread details to the system developers and framework developers. There is a push in academia for formalizing parellel patterns. As mentioned you cant always avoid threading, but there are an increasing number of tools in your arsenal to make it as painless as possible.
If you really want to live without threads, you can, so long as you don't call any functions that can potentially block. This may not be possible.
One alternative is to implement the tasks you would have made into threads as finite state machines. Basically, the task does what it can do immediately, then goes to its next state, waiting for an event, such as input arriving on a file or a timer going off. X Windows, as well as most GUI toolkits, support this style. When something happens, they call a callback, which does what it needs to do and returns. For a FSM, the callback checks to see what state the task is in and what the event is to determine what to do immediately and what the next state will be.
Say you have an app that needs to accept socket connections, and for each connection, parse command lines, execute some code, and return the results. A task would then be what listens to a socket. When select() (or Gtk+, or whatever) tells you the socket has something to read, you read it into a buffer, then check to see if you have enough input buffered to do something. If so, you advance to a "start doing something" state, otherwise you stay in the "reading a line" state. (What you "do" could be multiple states.) When done, your task drops the line from the buffer and goes back to the "reading a line" state. No threads or preemption needed.
This lets you act multithreaded by way of being event-driven. If your state machines are complicated, however, your code can get hard to maintain pretty fast, and you'll need to work up some kind of FSM-management library to separate the grunt work of running the FSM from the code that actually does things.
P.S. Another way to get threads without really using threads is the GNU Pth library. It doesn't do preemption, but it is another option if you really don't want to deal with threads.
Another approach to this may be to use a different concurrency model rather than avoid multithreading altogether (you have to utilize all these CPU cores in parallel somehow).
Take a look at mechanisms used in Clojure (e.g. agents, software transactional memory).
Software Transactional Memory (STM) is a good alternative concurrency control. It scales well with multiple processors and do not have most of the problems of conventional concurrency control mechanisms. It is implemented as part of the Haskell language. It worths giving a try. Although, I do not know how this is applicable in the context of SQLite.
Alternatives to threads:
coroutines
goroutines
mapreduce
workerpool
apple's grand central dispatch+lambdas
openCL
erlang
(interesting to note that half of those technologies were invented or popularised by google.)
Another thing is many web frameworks transparently use multiple threads/processes for handling requests, and usually in such a way that mostly eliminates the problems associated with multithreading (for the user of the framework), or at least makes the threading rather invisible. The web being stateless, the only shared state is session state (which isn't really a problem since by definition, a single session isn't going to be doing concurrent things), and data in a database that already has its multithreading nonsense sorted out for you.
It's somewhat important to note though that these are all abstractions. The underlying implementations of these things still use threads. But this is still incredibly useful. In the same way you wouldn't use assembler to write a web application, you wouldn't use threads directly to write any important application. Designing an application to use threads is too complicated to leave for a human to deal with.
Threading is not the only model of concurrency. The actors model (Erlang, Scala) is an example of a somewhat different approach.
http://www.scala-lang.org/node/242
If your task is really, really easily isolatable, you can use processes instead of threads, like Chrome does for its tabs.
Otherwise, inside a single process, there is no way to achieve real parallelism without threads, because you need at least two coroutines if you want two things to happen at the same time (assuming you're having multiple processors/cores at hand, of course; otherwise real parallelism is simply not possible).
The complexity of threading a program is always relative to the degree of isolation of the tasks the threads will perform. There's no trouble in running several threads if you know for sure these will never use the same variables. Then again, multiple high-level constructs exist in modern languages to help synchronize access to shared resources.
It's really a matter of application. If your task is simple enough to fit in some kind of high-level Task object (depends on your development platform; your mileage may vary), then using a task queue is your best bet. My rule of the thumb is that if you can't find a cool name to your thread, then its task is not important enough to justify a thread (instead of task going on an operation queue).
Threads give you the opportunity to do some evil things, specifically sharing state among different execution paths. But they offer a lot of convenience; you don't have to do expensive communication across process boundaries. Plus, they come with less overhead. So I think they're perfectly fine, used correctly.
I think the key is to share as little data as possible among the threads; just stick to synchronization data. If you try to share more than that, you have to engage in complex code that is hard to get right the first time around.
One method of avoiding threads is multiplexing - in essence you make a lightweight mechanism similar to threads which you manage yourself.
Thing is this is not always viable. In your case the 30s polling time per website - can it be split into 60 0.5s pieces, in between which you can stuff calls to the UI? If not, sorry.
Threads aren't evil, they are just easy to shoot your foot with. If doing Query A takes 30s and then doing Query B takes another 30s, doing them simultaneously in threads will take 120s instead of 60 due to thread overhead, fighting for disk access and various bottlenecks.
But if Operation A consists of 5s of activity and 55 seconds of waiting, mixed randomly, and Operation B takes 60s of actual work, doing them in threads will take maybe 70s, compared to plain 120 when you execute them in sequence.
The rule of thumb is: threads should idle and wait most of the time. They are good for I/O, slow reads, low-priority work and so on. If you want performance, use multiplexing, which requires more work but is faster, more efficient and has way less caveats. (synchronizing threads and avoiding race conditions is a whole different chapter of thread headaches...)

Why don't large programs (such as games) use loads of different threads? [closed]

As it currently stands, this question is not a good fit for our Q&A format. We expect answers to be supported by facts, references, or expertise, but this question will likely solicit debate, arguments, polling, or extended discussion. If you feel that this question can be improved and possibly reopened, visit the help center for guidance.
Closed 9 years ago.
I don't know how commercial games work inside very much, but the open source games I have come across don't seem to be massively into threading. Same goes for most other desktop applications, normally two or three threads seem to be used (eg program logic and GUI updates).
Why don't games have many threads? Eg separate threads for physics, sound, graphics, AI etc?
I don't know about the games that you have played, but most games run the sound on a separate thread. Networking code, at least the socket listeners run on a separate thread.
However, the rest of the game engine generally runs in a single thread. There are reasons for this. For example, most processing in a game runs a single chain of dependencies. Graphics depend on state of physics engine as does the artificial intelligence. Designing for multiple threads means that you have to have frame latency between the various subsystems for concurrency. You get quicker response time and snappier game play if these subsystems are computed linearly each frame. The part of the game that benefits the most from parallelization is of course the rendering subsystem which is offloaded to highly parallelized graphics accelerator cards.
You need to think, what are the actual benefits of threads? Remember that on a single core machine, threads don't actually allow concurrent execution, just the impression of it. Behind the scenes, the CPU is context-switching between the different threads, doing a little work on each every time. Therefore, if I have several tasks that involve no waiting, running them concurrently (on a single core) will be no quicker than running them linearly. In fact, it will be slower, due to the added overhead of the frequent context-switching.
If that is the case then, why ever use threads on a single core machine? Well firstly, because sometimes tasks can involve long periods of waiting on some external resource, such as a disk or other hardware device, to become available. Whilst the task in a waiting stage, threading allows other tasks to continue, thus using the CPU's time more efficiency.
Secondly, tasks may have a deadline of some sort in which to complete, particularly if they are responding to an event. The classic example is the user interface of an application. The computer should respond to user action events as quickly as possible, even if it is busy performing some other long running task, otherwise the user will be become agitated and may believe the application has crashed. Threading allows this to happen.
As for games, I am not a games programmer, but my understanding of the situation is this: 3D games create a programmatic model of the game world; players, enemies, items, terrain, etc. This game world is updated in discrete steps, based on the amount of time that has elapsed since the previous update. So, if 1ms has passed since the last time round the game loop, the position of an object is updated by using its velocity and the elapsed time to determine the delta (obviously the physics is a bit more complicated than that, but you get the idea). Other factors such as AI and input keys may also contribute to the update. When everything is finished, the updated game world is rendered as a new frame and the process begins again. This process usually occurs many times per second.
When we think about the game loop in this way, we can see that the engine is in fact achieving a very similar goal to that of threading. It has a number of long running tasks (updating the world's physics, handling user input, etc), and it gives the impression that they are happening concurrently by breaking them down into small pieces of work and interleaving these pieces, but instead of relying on the CPU or operating system to manage the time spent on each, it is doing it itself. This means it can keep all the different tasks properly synchronized, and avoid the complexities that come with real threading: locks, pre-emption, re-entrant code, etc. There is no performance implication to this approach either, because as we said a single core machine can only really execute code linearly anyway.
Things change when have a multi-core system. Now, tasks can be running genuinely concurrently and there may indeed be a benefit to using threading to handle different parts of the game world updates, so long as we can manage to synchronise the results to render consistent frames. We would expect therefore, that with the advent of multi-core systems, games engine developers would be working on this. And so it turns out, they are. Valve, the makers of Half Life, have recently introduced multi-processor support into their Source Engine, and I imagine many other engine developers are following suit.
Well, that turned out a little longer than I expected. I'm not a threading or games expert, but I hope I haven't made any especially glaring errors. If I have I'm sure people will correct me :)
The main reason is that, as elegant as it sounds, using multiple threads in a program as complicated as a 3D game is really, really, really difficult. Also, before the fairly recent introduction of low cost multi-core systems, using multiple threads did not offer much of a performance incentive.
Many games these days are using "task" or "job" systems for parallel processing. That is, the game spawns a fixed number of worker threads which are used for multiple tasks. Work is divided up into small pieces and queued, then sent to be processed by the worker threads as they become available.
This is becoming especially common on consoles. The PS3 is based on Cell architecture so you need to use parallel processing to get the best performance out of the system. The Xbox 360 can emulate a task/job setup that was designed for PS3 as it has multiple cores. You would probably find for most games that a lot of the system design is shared among the 360, PS3, and PC codebases, so PC most likely uses the same sort of tactic.
While it is hard to write threadsafe code, as many of the other answers indicate, I think there are a few other reasons for the things you're seeing:
First, many open source games are a few years old. Especially with this generation of consoles parallel programming is becoming popular and even necessary as mentioned above.
Second, very few open source projects seem concerned about getting the highest possible performance. As John Carmack pointed out to the Utah GLX project, highly optimized code is often harder to maintain than unoptimized code, so the latter would generally be preferred in open source contexts.
Third, I wouldn't take a small number of threads created by a game to mean that it's not using parallel jobs well.
I was about to post the same thing as William, but I'd like to expand on it a little bit. It's very hard to write optimal code for the future. Given the choice between writing something that will scale to hardware you don't have vs. writing something that will work on hardware you do have, most people will chose to do the latter. Since the single-core paradigm has been with us for so long, most code that has been written (especially for games where there is extreme pressure to get it out the door) isn't that future proof.
x86 has been very kind to game programmers, since we haven't had to think about the ramifications of less forgiving hardware platforms.
The fact that everybody here is correctly claiming that multithreading is hard is very sad. We desperately need to make concurrency systems easy.
Personally I think we are going to need a paradigm shift and new tools.
Other than the technical challenges of programming for multiple cores, commercial games have to run well on low end systems w/o multiple cores to make money.
Now that multi-core processors have been out for a while and the major game consoles have multiple cores it's only a matter of time before dual core shows up on the minimum system requirements list for PC games.
Here's a link to an interview with Orion Granatir from Intel where he's talking about getting game developers to take advantage of multi-threading.
There are many issues with race conditions and data locking when using lots of threads. Since the different parts of games are fairly reliant on each other it doesn't make much sense to do all the extra engineering required to use loads of threads.
It's very difficult to use threads without problems, and most GUI APIs are based on event driven coding anyway. Threads mandate the use of locking mechanisms which add delay to the code, and often that delay is unpredictable depending on who is currently holding the lock.
It seems sensible to me to have a single (or perhaps very few) threads handling things in an event driven way rather than hundreds of threads all causing strange and unrepeatable bugs.
Threads are dead, baby.
Realistically, in game development, threads don't scale beyond offloading very dedicated tasks like networking and loading. Job-systems seem to be the only way forward, given 8 CPU systems are becoming more commonplace even on PCs. And you can pretty much guarantee that upcoming super-multicore systems like Intel's Larrabee will be job-system based.
This has been a somewhat painful realization on Playstation3 and XBOX360 projects, and it seems now even Apple has jumped on board with their "revolutionary" Grand Central Dispatch system in Snow Leopard.
Threads have their place, but the naive promise of "put everything in a thread and it will all run faster" simply doesn't work in practice.

Best programming approach/methodology to assure thread safety

When I was learning Java coming from a background of some 20 years of procedural programming with basic, Pascal, COBOL and C, I thought at the time that the hardest thing about it was wrapping my head around the OOP jargon and concepts. Now with about 8 years of solid Java under my belt, I have come to the conclusion that the single hardest thing about programming in Java and similar languages like C# is the multithreaded/concurrent aspects.
Coding reliable and scalable multi-threaded applications is just plain hard! And with the trend for processors to grow "wider" rather than faster, it is rapidly becoming just plain critical.
The hardest area is, of course, controlling interactions between threads and the resulting bugs: deadlocks, race conditions, stale data and latency.
So my question to you is this: what approach or methodology do you employ for producing safe concurrent code while mitigating the potential for deadlocks, latency, and other problems? I have come up with an approach which is a little unconventional but has worked very well in several large applications, which I will share in a detailed answer to this question.
This not only applies to Java but to threaded programming in general. I find myself avoiding most of the concurrency and latency problems just by following these guidelines:
1/ Let each thread run its own lifetime (i.e., decide when to die). It can be prompted from outside (say a flag variable) but it in entirely responsible.
2/ Have all threads allocate and free their resources in the same order - this guarantees that deadlock will not happen.
3/ Lock resources for the shortest time possible.
4/ Pass responsibility for data with the data itself - once you notify a thread that the data is its to process, leave it alone until the responsibility is given back to you.
There are a number of techniques which are coming into the public consciousness just now (as in: the last few years). A big one would be actors. This is something that Erlang first brought to the grid iron but which has been carried forward by newer languages like Scala (actors on the JVM). While it is true that actors don't solve every problem, they do make it much easier to reason about your code and identify trouble spots. They also make it much simpler to design parallel algorithms because of the way they force you to use continuation passing over shared mutable state.
Fork/Join is something you should look at, especially if you're on the JVM. Doug Lea wrote the seminal paper on the topic, but many researchers have discussed it over the years. As I understand it, Doug Lea's reference framework is scheduled for inclusion into Java 7.
On a slightly less-invasive level, often the only steps necessary to simplify a multi-threaded application are just to reduce the complexity of the locking. Fine-grained locking (in the Java 5 style) is great for throughput, but very very difficult to get right. One alternative approach to locking which is gaining some traction through Clojure would be software-transactional memory (STM). This is essentially the opposite of conventional locking in that it is optimistic rather than pessimistic. You start out by assuming that you won't have any collisions, and then allow the framework to fix the problems if and when they occur. Databases often work this way. It's great for throughput on systems with low collision rates, but the big win is in the logical componentization of your algorithms. Rather than arbitrarily associating a lock (or a series of locks) with some data, you just wrap the dangerous code in a transaction and let the framework figure out the rest. You can even get a fair bit of compile-time checking out of decent STM implementations like GHC's STM monad or my experimental Scala STM.
There are a lot of new options for building concurrent applications, which one you pick depends greatly on your expertise, your language and what sort of problem you're trying to model. As a general rule, I think actors coupled with persistent, immutable data structures are a solid bet, but as I said, STM is a little less invasive and can sometimes yield more immediate improvements.
Avoid sharing data between threads where possible (copy everything).
Never have locks on method calls to external objects, where possible.
Keep locks for the shortest amount of time possible.
There is no One True Answer for thread safety in Java. However, there is at least one really great book: Java Concurrency in Practice. I refer to it regularly (especially the online Safari version when I'm on travel).
I strongly recommend that you peruse this book in depth. You may find that the costs and benefits of your unconventional approach are examined in depth.
I typically follow an Erlang style approach. I use the Active Object Pattern.
It works as follows.
Divide your application into very coarse grained units. In one of my current applications (400.000 LOC) I have appr. 8 of these coarse grained units. These units share no data at all. Every unit keeps its own local data. Every unit runs on its own thread (= Active Object Pattern) and hence is single threaded. You don't need any locks within the units. When the units need to send messages to other units they do it by posting a message to a queue of the other units. The other unit picks the message from the queue and reacts on that message. This might trigger other messages to other units.
Consequently the only locks in this type of application are around the queues (one queue and lock per unit). This architecture is deadlock free by definition!
This architecture scales extremely well and is very easy to implement and extend as soon as you understood the basic principle. It like to think of it as a SOA within an application.
By dividing your app into the units remember. The optimum number of long running threads per CPU core is 1.
I recommend flow-based programming, aka dataflow programming. It uses OOP and threads, I feel it like a natural step forward, like OOP was to procedural. Have to say, dataflow programming can't be used for everything, it is not generic.
Wikipedia has good articeles on the topic:
http://en.wikipedia.org/wiki/Dataflow_programming
http://en.wikipedia.org/wiki/Flow-based_programming
Also, it has several advantages, as the incredible flexibile configuration, layering; the programmer (Component programmer) has not to program the business logic, it's done in another stage (putting the processing network together).
Did you know, make is a dataflow system? See make -j, especially if you have multi-core processor.
Writing all the code in a multi-threaded application very... carefully! I don't know any better answer than that. (This involves stuff like jonnii mentioned).
I've heard people argue (and agree with them) that the traditional threading model really won't work going into the future, so we're going to have to develop a different set of paradigms / languages to really use these newfangled multi-cores effectively. Languages like Haskell, whose programs are easily parallelizable since any function that has side effects must be explicitly marked that way, and Erlang, which I unfortunately don't know that much about.
I suggest the actor model.
The actor model is what you are using and it is by far the simplest (and efficient way) for multithreading stuff. Basically each thread has a (synchronized) queue (it can be OS dependent or not) and other threads generate messages and put them in the queue of the thread that will handle the message.
Basic example:
thread1_proc() {
msg = get_queue1_msg(); // block until message is put to queue1
threat1_msg(msg);
}
thread2_proc() {
msg = create_msg_for_thread1();
send_to_queue1(msg);
}
It is a tipical example of producer consumer problem.
It is clearly a difficult problem. Apart from the obvious need for carefulness, I believe that the very first step is to define precisely what threads you need and why.
Design threads as you would design classes : making sure you know what makes them consistent : their contents and their interactions with other threads.
I recall being somewhat shocked to discover that Java's synchronizedList class wasn't fully thread-safe, but only conditionally thread-safe. I could still get burned if I didn't wrap my accesses (iterators, setters, etc.) in a synchronized block. This means that I might've assured my team and my management that my code was thread safe, but I might've been wrong. Another way I can assure thread safety is for a tool to analyse the code and have it pass. STP, Actor model, Erlang, etc are some ways of getting the latter form of assurance. Being able to assure properties of a program reliably is/will be a huge step forward in programming.
Looks like your IOC is somewhat FBP-like :-) It would be fantastic if the JavaFBP code could get a thorough vetting from someone like yourself versed in the art of writing thread-safe code... It's on SVN in SourceForge.
Some experts feel the answer to your question is to avoid threads altogether, because it's almost impossible to avoid unforseen problems. To quote The Problem with Threads:
We developed a process that included
a code maturity rating system (with four levels, red, yellow, green, and blue), design reviews, code
reviews, nightly builds, regression tests, and automated code coverage metrics. The portion
of the kernel that ensured a consistent view of the program structure was written in early 2000,
design reviewed to yellow, and code reviewed to green. The reviewers included concurrency experts,
not just inexperienced graduate students (Christopher Hylands (now Brooks), Bart Kienhuis, John
Reekie, and [Ed Lee] were all reviewers). We wrote regression tests that achieved 100 percent code
coverage...
The... system itself began to be widely used, and every use of the system exercised this
code. No problems were observed until the code deadlocked on April 26, 2004, four years later.
The safest approach to design new applications with multi threading is to adhere to the rule:
No design below the design.
What does that mean?
Imagine you identified major building blocks of your application. Let it be the GUI, some computations engines. Typically, once you have a large enough team size, some people in the team will ask for "libraries" to "share code" between those major building blocks. While it was relatively easy in the start to define the threading and collaboration rules for the major building blocks, all that effort is now in danger as the "code reuse libraries" will be badly designed, designed when needed and littered with locks and mutexes which "feel right".
Those ad-hoc libraries are the design below your design and the major risk for your threading architecture.
What to do about it?
Tell them that you rather have code duplication than shared code across thread boundaries.
If you think, the project will really benefit from some libraries, establish the rule that they must be state-free and reentrant.
Your design is evolving and some of that "common code" could be "moved up" in the design to become a new major building block of your application.
Stay away from the cool-library-on-the-web-mania. Some third party libraries can really save you a lot of time. But there is also a tendency that anyone has their "favorites", which are hardly essential. And with each third party library you add, your risk of running into threading problems increases.
Last not least, consider to have some message based interaction between your major building blocks; see the often mentioned actor model, for example.
The core concerns as I saw them were (a) avoiding deadlocks and (b) exchanging data between threads. A lessor concern (but only slightly lessor) was avoiding bottlenecks. I had already encountered several problems with disparate out of sequence locking causing deadlocks - it's very well to say "always acquire locks in the same order", but in a medium to large system it is practically speaking often impossible to ensure this.
Caveat: When I came up with this solution I had to target Java 1.1 (so the concurrency package was not yet a twinkle in Doug Lea's eye) - the tools at hand were entirely synchronized and wait/notify. I drew on experience writing a complex multi-process communications system using the real-time message based system QNX.
Based on my experience with QNX which had the deadlock concern, but avoided data-concurrency by coping messages from one process's memory space to anothers, I came up with a message-based approach for objects - which I called IOC, for inter-object coordination. At the inception I envisaged I might create all my objects like this, but in hindsight it turns out that they are only necessary at the major control points in a large application - the "interstate interchanges", if you will, not appropriate for every single "intersection" in the road system. That turns out to be a major benefit because they are quite un-POJO.
I envisaged a system where objects would not conceptually invoke synchronized methods, but instead would "send messages". Messages could be send/reply, where the sender waits while the message is processed and returns with the reply, or asynchronous where the message is dropped on a queue and dequeued and processed at a later stage. Note that this is a conceptual distinction - the messaging was implemented using synchronized method calls.
The core objects for the messaging system are an IsolatedObject, an IocBinding and an IocTarget.
The IsolatedObject is so called because it has no public methods; it is this that is extended in order to receive and process messages. Using reflection it is further enforced that child object has no public methods, nor any package or protected methods except those inherited from IsolatedObject nearly all of which are final; it looks very strange at first because when you subclass IsolatedObject, you create an object with 1 protected method:
Object processIocMessage(Object msgsdr, int msgidn, Object msgdta)
and all the rest of the methods are private methods to handle specific messages.
The IocTarget is a means of abstracting visibility of an IsolatedObject and is very useful for giving another object a self-reference for sending signals back to you, without exposing your actual object reference.
And the IocBinding simply binds a sender object to a message receiver so that validation checks are not incurred for every message sent, and is created using an IocTarget.
All interaction with the isolated objects is through "sending" it messages - the receiver's processIocMessage method is synchronized which ensures that only one message is be handled at a time.
Object iocMessage(int mid, Object dta)
void iocSignal (int mid, Object dta)
Having created a situation where all work done by the isolated object is funneled through a single method, I next arranged the objects in a declared hierarchy by means of a "classification" they declare when constructed - simply a string that identifies them as being one of any number of "types of message receiver", which places the object within some predetermined hierarchy. Then I used the message delivery code to ensure that if the sender was itself an IsolatedObject that for synchronous send/reply messages it was one which is lower on the hierarchy. Asynchronous messages (signals) are dispatched to message receivers using separate threads in a thread pool who's entire job deliver signals, therefore signals can be send from any object to any receiver in the system. Signals can can deliver any message data desired, but not reply is possible.
Because messages can only be delivered in an upward direction (and signals are always upward because they are delivered by a separate thread running solely for that purpose) deadlocks are eliminated by design.
Because interactions between threads are accomplished by exchanging messages using Java synchronization, race conditions and issues of stale data are likewise eliminated by design.
Because any given receiver handles only one message at a time, and because it has no other entry points, all considerations of object state are eliminated - effectively, the object is fully synchronized and synchronization cannot accidentally be left off any method; no getters returning stale cached thread data and no setters changing object state while another method is acting on it.
Because only the interactions between major components is funneled through this mechanism, in practice this has scaled very well - those interactions don't happen nearly as often in practice as I theorized.
The entire design becomes one of an orderly collection of subsystems interacting in a tightly controlled manner.
Note this is not used for simpler situations where worker threads using more conventional thread pools will suffice (though I will often inject the worker's results back into the main system by sending an IOC message). Nor is it used for situations where a thread goes off and does something completely independent of the rest of the system such as an HTTP server thread. Lastly, it is not used for situations where there is a resource coordinator that itself does not interact with other objects and where internal synchronization will do the job without risk of deadlock.
EDIT: I should have stated that the messages exchanged should generally be immutable objects; if using mutable objects the act of sending it should be considered a hand over and cause the sender to relinquish all control, and preferably retain no references to the data. Personally, I use a lockable data structure which is locked by the IOC code and therefore becomes immutable on sending (the lock flag is volatile).

switch to parallel coding

we all writing code for single processor.
i wonder when we all are able to write code on multi processors?
what do we need (software tools, logic, algorithms) for this switching?
edit: in my view, as we do many task parallely, same way we need to convert those real life solutions(algorithms) to computer lang. just as OOPs coding did for procedural coding. OOPs is more real life coding style than procedural one. so i hope for that kind of solutions.
I think the most important requirement is a good language that has native constructs that support parallelism or one that can automatically generate parallel code. There are quite a few languages that fit that description, but none of them is popular enough to really be considered for mainstream use. That, in turn is caused by several things:
By their very nature, these languages are very different from today's imperative languages, and are therefor harder to learn (or at least seem that way).
They often lack good tools and libraries, making them unusable for any "real" project.
Of course, if it were more popular more people would be willing to learn it and there would be more support, so it's a kind of cycle that's pretty hard to break out of. I guess all we can do is hope. :)
An example of a language designed with heavy parallelization in mind is Erlang - and it's actually used in commercial projects.
What we need are natural abstractions for highly-concurrent algorithms. Actors (think: Erlang) go a long way in this direction, but they aren't a one-size-fits-all solution. Some more specific abstractions like fork/join or map/reduce can be even easier to apply to common problems.
The trick with all of these concurrency abstractions is they require functional-style programming. Concurrency doesn't mesh well with shared mutable state. As they say, "Locks considered harmful". Since most developers come from a strictly imperative background, switching to a shared-nothing continuation passing approach is often extremely challenging.
Incidentally, with respect to concurrency abstractions, Clojure has some very interesting features in this direction. Not only does it have sort-of actors, but it also defines a transactional memory model (think: databases) along with a global, atomic references mechanism. These two features allow concurrent operations to share "mutable" state without ever having to worry about locking or race conditions.
In the end, it comes down to education. Much of the needed theoretical work into concurrency abstractions has already been done, we just need to accept it. Unfortunately, as Erlang and Haskell prove, sometimes the best ideas remain relegated to an extremely fringe demographic. Hopefully efforts like Scala and Clojure will succeed in bringing the more advanced abstractions into the mainstream by sneaking them onto an existing, well-supported platform (the JVM).
Unfortunately for massive concurrent programming - unless there is a breakthrough in compilers to help, we will be throwing out a lot of what we know about algorithms (I think Don Knuth even said that). Read about Erlang for a glimpse of this possible future.
There are several tools/languages that are popular or are gaining popularity. If you use FORTRAN, C, or C++, you can use OpenMP (not too hard to implement) or the Message Passing Interface (MPI) libraries (powerful and greatest speedup potential, but also complex and difficult). OpenMP uses preprocessor directives to mark areas that can be parallelized, especially loops. MPI uses messages that pass data back and forth between processes, and the greatest difficulty is keeping everything synchronized without hitting bottlenecks and keeping processes waiting. I would say MPI is definitely on the way out, however. It's become clear in the scientific/high-performance computing communities that the speedup is rarely worth the additional development time.
As for up and coming languages, check out Fortress. It's still being designed, but the goal is to create a language even easier for scientific computing than FORTRAN. Programs will be specified in a very high level mathematical syntax. Additionally, parallelism will be implicit; the programmer will have to work to do things in serial. Plus, it's being championed by Sun and is based on java, so it will be portable.
There is no simple answer, and in many ways even the complex answers are currently inadequate or incomplete. You'll get a better answer if you are more specific about the replies you want: pointers to dev libraries and tools, instructional materials, pointers to current research projects and issues in this area, or something else?
The most important requirement is to be able to split your problem into smaller problems that can be solved independently of each other. Once you've worked out how you're going to do that, everything else is easier to think about and further questions of implementation (e.g. "parts of my calculation depend on other parts - how do I wait for them to have finished?") become concrete, specific things you can research or ask here about.
for java you can now look to Parallel Java Library or DPJ(deterministic Parallel Java!)
It will offer you great help in extracting parallelism from codes!!

Resources