Develop Reference

Volatile.Write freshness guarentee

The documentation for Volatile.Write says the following:
Writes the specified object reference to the specified field. On
systems that require it, inserts a memory barrier that prevents the
processor from reordering memory operations as follows: If a read or
write appears before this method in the code, the processor cannot
move it after this method.
and
value T
The object reference to write. The reference is written
immediately so that it is visible to all processors in the computer.
But it seems like quotes 1 and 2 are contradictory.
For the second quote to be true, I would think that the first quote would have to be changed as follows:
If a read or
write appears before after this method in the code, the processor cannot
move it after before this method.
Does Volatile.Write actually mean that other threads are guaranteed to pick up the write in a timely fashion, or is the second quote misleading?
It seems to me as though all these "Volatile"/"Memory Barriers" seem to be focused on is ensuring that if writes are exposed to other threads they are exposed in the correct order, but I can't seem to find what actually would be force them to be exposed.
I understand that it may be hard/impossible to expose writes to other threads immediately, but without volatile writes/reads there are cases when the writes are exposed never. So it seems there must be a way to ensure that writes are exposed "eventually", but I'm unsure what that is. Is it that writes are always exposed in .NET but reads can be cached? And if so does Volatile.Read stop this caching behaviour?
(Note I have read through Joseph Albahari's Threading in C# which tends to suggest I need explicit memory barriers before my reads and after my writes, although it's not clear why even that should be effective as the documentation for Thread.MemoryBarrier doesn't seem to explicitly say that the writes are shown to other threads).

You are misunderstanding the concept of barriers a little bit. As you wrote
The object reference to write. The reference is written immediately so that it is visible to all processors in the computer.
So the really important unit here is a processor, not thread.
So, there are processors, processor caches, store buffers and invalidation queues involved.
When a processor writes something into the memory, it looks like that or similar to that
The subject is at the store buffer level. As you can see, there are a lot of things is going on when you write something or read, and it does not happen instantly for all the processors in the system. At the beginning a read or write command is places into processor's store buffer, and those commands could be reordered, in other words, executed in different order by the processor.
While that happens, other processors don't know about changes, if the operation is write and the currently working processor doesn't know about changes other processors made.
When you place a barrier, that means that operations in the store buffer or invalidation queue should be completed before any read or write could be performed. That is necessary to actualize CPU caches across processors. So there is basically no mechanics to synchronize any data across threads, we are syncing data across processors.
When a thread A writes something on processor 1 and thread B reads something on the processor 1, they both starts by looking into the store buffer first, so they read actual data, whether any barriers placed or not.
It's just an overview of the mechanic involved, maybe I'm wrong in some details. You can find complete info if you read about MESI protocol, this PDF with explanation on invalidation queues and store buffers

I agree with you that the description in the MSDN documentation is bit confusing. I would say that "immediately" is strong word here as well as in regards to any subject related to parallel processes. The result won't be visible immediately but documentation doesn't say that - it says that the value will be written immediately, that is as soon as all prior load/store operation results become globally visible the store operation to write a value will be immediately initiated.
As for the memory barriers, they only can give a guarantee of prior operations exposure (global visibility) because in essence the memory barriers are instructions which are encountered by a CPU make the CPU "wait" for getting all pending load/store operations globally visible while the moment of global visibility of value written by Volatile.Write is neither barrier nor Volatile.Write concern.
Now about suggestion to use the barrier in lock-free programming. Of course it makes sense because it ensures the order of global visibility which is actual for multi-core systems. When you cannot be sure that an event B always happens after event A you just can't build reliable logic supposed to be executed in multi-core environemnts.

guarantee that up-to-date value of variable is always visible to several threads on multi-processor system

I'm using such configuration:
.NET framework 4.5
Windows Server 2008 R2
HP DL360p Gen8 (2 * Xeon E5-2640, x64)
I have such field somewhere in my program:
protected int HedgeVolume;
I access this field from several threads. I assume that as I have multi-processor system it's possible that this threads are executing on different processors.
What should I do to guarantee that any time I use this field the most recent value is "read"? And to make sure that when I "write" value it become available to all other threads immediately?
What should I do?
just leave field as is.
declare it volatile
use Interlocked class to access the field
use .NET 4.5 Volatile.Read, Volatile.Write methods to access the field
use lock
I only need simplest way to make my program work on this configuration I don't need my program to work on another computers or servers or operation systems. Also I want minimal latency so I'm looking for fastest solution that will always work on this standard configuration (multiprocessor intel x64, .net 4.5).

Your question is missing one key element... How important is the integrity of the data in that field?
volatile gives you performance, but if a thread is currently writing changes to the field, you won't get that data until it's done, so you might access out of date information, and potentially overwrite changes another thread is currently doing. If the data is sensitive, you might get bugs that would get very hard to track. However, if you are doing very quick update, overwrite the value without reading it and don't care that once in a while you get outdated (by a few ms) data, go for it.
lock guaranty that only one thread can access the field at a time. You can put it only on the methods that write the field and leave the reading method alone. The down side is, it is slow, and may block a thread while another is performing its task. However, you are sure your data stay valid.
Interlock exist to shield yourself from the scheduler context switch. My opinion? Don't use it unless you know exactly why you would be using it and exactly how to use it. It gives options, but with great options comes great problematic. It prevents a context switch while a variable is being update. It might not do what you think it does and won't prevent parallel threads from performing their tasks simultaneously.

You want to use Volatile.Read().
As you are running on x86, all writes in C# are the equivalent of Volatile.Write(), you only need to use this for Itanium.
Volatile.Read() will ensure that you get the latest copy regardless of which thread last wrote it.
There is a fantastic write up here, C# Memory Model Explained
Summary of it includes,
On some processors, not only must the compiler avoid certain
optimizations on volatile reads and writes, it also has to use special
instructions. On a multi-core machine, different cores have different
caches. The processors may not bother to keep those caches coherent by
default, and special instructions may be needed to flush and refresh
the caches.
Hopefully that much is obvious, other than the need for volatile to stop the compiler from optimising it, there is the processor as well.
However, in C# all writes are volatile (unlike say in Java),
regardless of whether you write to a volatile or a non-volatile field.
So, the above situation actually never happens in C#. A volatile write
updates the thread’s cache, and then flushes the entire cache to main
memory.
You do not need Volatile.Write(). More authoratitive source here, Joe Duffy CLR Memory Model. However, you may need it to stop the compiler reordering it.
Since all C# writes are volatile, you can think of all writes as going
straight to main memory. A regular, non-volatile read can read the
value from the thread’s cache, rather than from main
You need Volatile.Read()

When you start designing a concurrent program, you should consider these options in order of preference:
1) Isolation: each thread has it's own private data
2) Immutability: threads can see shared state, but it never changes
3) Mutable shared state: protect all access to shared state with locks
If you get to (3), then how fast do you actually need this to be?
Acquiring an uncontested lock takes in the order of 10ns ( 10-8 seconds ) - that's fast enough for most applications and is the easiest way to guarantee correctness.
Using any of the other options you mention takes you into the realm of low-lock programming, which is insanely difficult to get correct.
If you want to learn how to write concurrent software, you should read these:
Intro: Joe Albahari's free e-book - will take about a day to read
Bible: Joe Duffy's "Concurrent Programming on Windows" - will take about a month to read

Depends what you DO. For reading only, volatile is easiest, interlocked allows a little more control. Lock is unnecessary as it is more ganular than the problem you describe. Not sure about Volatile.Read/Write, never used them.

volatile - bad, there are some issues (see Joe Duffy's blog)
if all you do is read the value or unconditionally write a value - use Volatile.Read and Volatile.Write
if you need to read and subsequently write an updated value - use the lock syntax. You can however achieve the same effect without lock using the Interlocked classes functionality, but this is more complex (involves CompareExchange s to ensure that you are updating the read value i.e. has not been modified since the read operation + logic to retry if the value was modified since the read).

From this i can understand that you want to be able to read the last value that it was writtent in a field. Lets make an analogy with the sql concurency problem of the data. If you want to be able to read the last value of a field you must make atomic instructions. If someone is writing a field all of the threads must be locked for reading until that thread finished the writing transaction. After that every read on that thread will be safe. The problem is not with reading as it is with writing. A lock on that field whenever its writtent should be enough if you ask me ...

First have a look here: Volatile vs. Interlocked vs. lock
The volatile modifier shurely is a good option for a multikernel cpu.
But is this enough? It depends on how you calculate the new HedgeVolume value!
If your new HedgeVolume does not depend on current HedgeVolume then your done with volatile.
But if HedgeVolume[x] = f(HedgeVolume[x-1]) then you need some thread synchronisation to guarantee that HedgeVolume doesn't change while you calculate and assign the new value. Both, lock and Interlocked szenarios would be suitable in this case.

I had a similar question and found this article to be extremely helpful. It's a very long read, but I learned a LOT!

Whats the Best Concurrent Thread Shared Memory Architecture Without Locking?

I have a 2d Array of memory. I have multiple threads reading and writing to single elements in the array spontaneously, arbitrarily, and concurrently.
What is the fastest way or best practice to construct my memory access code? I don't like the idea of locking because it blocks other threads.
Data integrity is actually not that important, but it should be (mostly) consistent. My code can handle a few memory errors.
It needs to be really, really fast!
Thanks for feedback.

If data integrity is not important, you can just access the data without caring about multithreading at all.
No one can predict the result, though.
I wouldn't call this approach "best practice", however. IMHO best pratice is caring about multithreading, and protecting the data with appropriately-grained mutexes. My opinion is that every application should be first correct, and only then fast. Inconsistent results are just wrong, doesn't matter if they come fast or not.

Use the Interlocked class to CAS (CompareAndExchange) the objects/values in your array. It makes the operation atomic which ensures that the data is not corrupted. That's about the fastest thing you can do (aside from accessing/modifying the data directly without interlocking). However, if you're modifying the size of the 2D array (growing/shrinking) then you will have some serious problems unless you use some kind of locking mechanism on your array.

Declare the array as volatile and ensure it's scoped such that it's visible to all your threads. I generally like to avoid statics, so either pass the array by reference, or set up all your threads to run methods of an instance class that has the array defined as an instance field.
However, I strongly urge you to rethink what "volatile access" means in terms of data integrity. Best practice is NOT to do what you are attempting without good locking mechanics. You may think it's a small problem, but you can find yourself with a very non-deterministic system, so much so that its data isn't reliable in the slightest.
Let's say you have 8 threads running, and all of them will get a value from an index of the array, do some calculation, then add the result back to the index of the array. Thread 1 starts first and gets the value of the index, 0. Then threads 2-7 all start and get the same value. Thread 1 performs its calculation, gets the index again to ensure it has the "latest" value, then tries to update the value. However, other threads are waiting for that memory, and due to some scheduling implementation you know nothing about, in between Thread 1 getting the index (still zero) and writing its result, threads 2-7 have ALL written their values. Then Thread 1 writes its value, overwriting everything the other 7 threads have done. The other 7 threads, in turn, probably had similar "races" with each other such that the value overwritten by Thread 1 probably overwrote the results of half the threads anyway.
I guarantee you that this behavior is NOT what you want, no matter how much you think you can get away with it; it WILL cause data corruption, which WILL affect other areas of the system, and you WILL be forced to implement proper locking.

If you are interested solely in performance, then the way in which you order your memory accesses can play a big role. Spend an hour or so reading through the slides from Lecture 1 of MIT's Performance Engineering class. The other lectures may also be interesting to you (such as Lecture 6).
Basically, you can optimize your use of the cache to greatly improve performance, depending on your read/write patterns, given the workload you are using.
This should not stop you from doing something that is correct, however.

Can I avoid using locks for my seldomly-changing variable?

I've been reading Joe Duffy's book on Concurrent programming. I have kind of an academic question about lockless threading.
First: I know that lockless threading is fraught with peril (if you don't believe me, read the sections in the book about memory model)
Nevertheless, I have a question:
suppose I have an class with an int property on it.
The value referenced by this property will be read very frequently by multiple threads
It is extremely rare that the value will change, and when it does it will be a single thread that changes it.
If it does change while another operation that uses it is in flight, no one is going to lose a finger (the first thing anyone using it does is copy it to a local variable)
I could use locks (or a readerwriterlockslim to keep the reads concurrent).
I could mark the variable volatile (lots of examples where this is done)
However, even volatile can impose a performance hit.
What if I use VolatileWrite when it changes, and leave the access normal for reads. Something like this:
public class MyClass
{
private int _TheProperty;
internal int TheProperty
{
get { return _TheProperty; }
set { System.Threading.Thread.VolatileWrite(ref _TheProperty, value); }
}
}
I don't think that I would ever try this in real life, but I'm curious about the answer (more than anything, as a checkpoint of whether I understand the memory model stuff I've been reading).

Marking a variable as "volatile" has two effects.
1) Reads and writes have acquire and release semantics, so that reads and writes of other memory locations will not "move forwards and backwards in time" with respect to reads and writes of this memory location. (This is a simplification, but you take my point.)
2) The code generated by the jitter will not "cache" a value that seems to logically be unchanging.
Whether the former point is relevant in your scenario, I don't know; you've only described one memory location. Whether or not it is important that you have only volatile writes but not volatile reads is something that is up to you to decide.
But it seems to me that the latter point is quite relevant. If you have a spin lock on a non-volatile variable:
while(this.prop == 0) {}
the jitter is within its rights to generate this code as though you'd written
if (this.prop == 0) { while (true) {} }
Whether it actually does so or not, I don't know, but it has the right to. If what you want is for the code to actually re-check the property on each go round the loop, marking it as volatile is the right way to go.

The question is whether the reading thread will ever see the change. It's not just a matter of whether it sees it immediately.
Frankly I've given up on trying to understand volatility - I know it doesn't mean quite what I thought it used to... but I also know that with no kind of memory barrier on the reading thread, you could be reading the same old data forever.

The "performance hit" of volatile is because the compiler now generates code to actually check the value instead of optimizing that away - in other words, you'll have to take that performance hit regardless of what you do.

At the CPU level, yes every processor will eventually see the change to the memory address. Even without locks or memory barriers. Locks and barriers would just ensure that it all happened in a relative ordering (w.r.t other instructions) such that it appeared correct to your program.
The problem isn't cache-coherency (I hope Joe Duffy's book doesn't make that mistake). The caches stay conherent - it is just that this takes time, and the processors don't bother to wait for that to happen - unless you enforce it. So instead, the processor moves on to the next instruction, which may or may not end up happening before the previous one (because each memory read/write make take a different amount of time. Ironically because of the time for the processors to agree on coherency, etc. - this causes some cachelines to be conherent faster than others (ie depending on whether the line was Modified, Exclusive, Shared, or Invalid it takes more or less work to get into the necessary state).)
So a read may appear old or from an out of date cache, but really it just happened earlier than expected (typically because of look-ahead and branch prediction). When it really was read, the cache was coherent, it has just changed since then. So the value wasn't old when you read it, but it is now when you need it. You just read it too soon. :-(
Or equivalently, it was written later than the logic of your code thought it would be written.
Or both.
Anyhow, if this was C/C++, even without locks/barriers, you would eventually get the updated values. (within a few hundred cycles typically, as memory takes about that long). In C/C++ you could use volatile (the weak non-thread volatile) to ensure that the value wasn't read from a register. (Now there's a non-coherent cache! ie the registers)
In C# I don't know enough about CLR to know how long a value could stay in a register, nor how to ensure you get a real re-read from memory. You've lost the 'weak' volatile.
I would suspect as long as the variable access doesn't completely get compiled away, you will eventually run out of registers (x86 doesn't have many to start with) and get your re-read.
But no guarantees that I see. If you could limit your volatile-read to a particular point in your code that was often, but not too often (ie start of next task in a while(things_to_do) loop) then that might be the best you can do.

This is the pattern I use when the 'last writer wins' pattern is applicable to the situation. I had used the volatile keyword, but after seeing this pattern in a code example from Jeffery Richter, I started using it.

For normal things (like memory-mapped devices), the cache-coherency protocols going on within/between the CPU/CPUs is there to ensure that different threads sharing that memory get a consistent view of things (i.e., if I change the value of a memory location in one CPU, it will be seen by other CPUs that have the memory in their caches). In this regard volatile will help to ensure that the optimizer doesn't optimize away memory accesses (which are always going through cache anyway) by, say, reading the value cached in a register. The C# documentation seems pretty clear on this. Again, the application programmer doesn't generally have to deal with cache-coherency themselves.
I highly recommend reading the freely available paper "What Every Programmer Should Know About Memory". A lot of magic goes on under the hood that mostly prevents shooting oneself in the foot.

In C#, the int type is thread-safe.
Since you said that only one thread writes to it, you should never have contention as to what is the proper value, and as long as you are caching a local copy, you should never get dirty data.
You may, however, want to declare it volatile if an OS thread will be doing the update.
Also keep in mind that some operations are not atomic, and can cause problems if you have more than one writer. For example, even though the bool type wont corrupt if you have more than one writer, a statement like this:
a = !a;
is not atomic. If two threads read at the same time, you have a race condition.

Spinlocks, How Useful Are They?

How often do you find yourself actually using spinlocks in your code? How common is it to come across a situation where using a busy loop actually outperforms the usage of locks?
Personally, when I write some sort of code that requires thread safety, I tend to benchmark it with different synchronization primitives, and as far as it goes, it seems like using locks gives better performance than using spinlocks. No matter for how little time I actually hold the lock, the amount of contention I receive when using spinlocks is far greater than the amount I get from using locks (of course, I run my tests on a multiprocessor machine).
I realize that it's more likely to come across a spinlock in "low-level" code, but I'm interested to know whether you find it useful in even a more high-level kind of programming?

It depends on what you're doing. In general application code, you'll want to avoid spinlocks.
In low-level stuff where you'll only hold the lock for a couple of instructions, and latency is important, a spinlock mat be a better solution than a lock. But those cases are rare, especially in the kind of applications where C# is typically used.

In C#, "Spin locks" have been, in my experience, almost always worse than taking a lock - it's a rare occurrence where spin locks will outperform a lock.
However, that's not always the case. .NET 4 is adding a System.Threading.SpinLock structure. This provides benefits in situations where a lock is held for a very short time, and being grabbed repeatedly. From the MSDN docs on Data Structures for Parallel Programming:
In scenarios where the wait for the lock is expected to be short, SpinLock offers better performance than other forms of locking.
Spin locks can outperform other locking mechanisms in cases where you're doing something like locking through a tree - if you're only having locks on each node for a very, very short period of time, they can out perform a traditional lock. I ran into this in a rendering engine with a multithreaded scene update, at one point - spin locks profiled out to outperform locking with Monitor.Enter.

For my realtime work, particularly with device drivers, I've used them a fair bit. It turns out that (when last I timed this) waiting for a sync object like a semaphore tied to a hardware interrupt chews up at least 20 microseconds, no matter how long it actually takes for the interrupt to occur. A single check of a memory-mapped hardware register, followed by a check to RDTSC (to allow for a time-out so you don't lock up the machine) is in the high nannosecond range (basicly down in the noise). For hardware-level handshaking that shouldn't take much time at all, it is really tough to beat a spinlock.

My 2c: If your updates satisfy some access criteria then they are good spinlock candidates:
fast, ie you will have time to acquire the spinlock, perform the updates and release the spinlock in a single thread quanta so that you don't get pre-empted while holding the spinlock
localized all data you update are in preferably one single page that is already loaded, you do not want a TLB miss while you holding the spinlock, and you definetely don't want an page fault swap read!
atomic you do not need any other lock to perform the operation, ie. never wait for locks under spinlock.
For anything that has any potential to yield, you should use a notified lock structure (events, mutex, semaphores etc).

One use case for spin locks is if you expect very low contention but are going to have a lot of them. If you don't need support for recursive locking, a spinlock can be implemented in a single byte, and if contention is very low then the CPU cycle waste is negligible.
For a practical use case, I often have arrays of thousands of elements, where updates to different elements of the array can safely happen in parallel. The odds of two threads trying to update the same element at the same time are very small (low contention) but I need one lock for every element (I'm going to have a lot of them). In these cases, I usually allocate an array of ubytes of the same size as the array I'm updating in parallel and implement spinlocks inline as (in the D programming language):
while(!atomicCasUbyte(spinLocks[i], 0, 1)) {}
myArray[i] = newVal;
atomicSetUbyte(spinLocks[i], 0);
On the other hand, if I had to use regular locks, I would have to allocate an array of pointers to Objects, and then allocate a Mutex object for each element of this array. In scenarios such as the one described above, this is just plain wasteful.

If you have performance critical code and you have determined that it needs to be faster than it currently is and you have determined that the critical factor is the lock speed, then it'd be a good idea to try a spinlock. In other cases, why bother? Normal locks are easier to use correctly.

Please note the following points :
Most mutexe's implementations spin for a little while before the thread is actually unscheduled. Because of this it is hard to compare theses mutexes with pure spinlocks.
Several threads spining "as fast as possible" on the same spinlock will consome all the bandwidth and drasticly decrease your program efficiency. You need to add tiny "sleeping" time by adding noop in your spining loop.

You hardly ever need to use spinlocks in application code, if anything you should avoid them.
I can't thing of any reason to use a spinlock in c# code running on a normal OS. Busy locks are mostly a waste on the application level - the spinning can cause you to use the entire cpu timeslice, vs a lock will immediatly cause a context switch if needed.
High performance code where you have nr of threads=nr of processors/cores might benefit in some cases, but if you need performance optimization at that level your likely making next gen 3D game, working on an embedded OS with poor synchronization primitives, creating an OS/driver or in any case not using c#.

I used spin locks for the stop-the-world phase of the garbage collector in my HLVM project because they are easy and that is a toy VM. However, spin locks can be counter-productive in that context:
One of the perf bugs in the Glasgow Haskell Compiler's garbage collector is so annoying that it has a name, the "last core slowdown". This is a direct consequence of their inappropriate use of spinlocks in their GC and is excacerbated on Linux due to its scheduler but, in fact, the effect can be observed whenever other programs are competing for CPU time.
The effect is clear on the second graph here and can be seen affecting more than just the last core here, where the Haskell program sees performance degradation beyond only 5 cores.

Always keep these points in your mind while using spinlocks:
Fast user mode execution.
Synchronizes threads within a single process, or multiple processes if in shared memory.
Does not return until the object is owned.
Does not support recursion.
Consumes 100% of CPU while "waiting".
I have personally seen so many deadlocks just because someone thought it will be a good idea to use spinlock.
Be very very careful while using spinlocks
(I can't emphasize this enough).