How computers work

[JohnOne]

Over the past few years I've had a general idea of how computers work, thanks to a lot on info on the internet.

It is my understanding that bits are passed around, instructions are given and carried out and all that electron jazz.

But I don't understand why race conditions for memory access are reached. I read an article recently in which the author mentioned that a CPU only carries out one instruction at a time, meaning linear, and that is regardless of how many cores it might have. I'm sorry I cannot remember the source of the article, it was some techno website, and I just considered it to be true because it made sense as I was reading it.

So if it is true, why ever does a race condition between processes or threads occur?

An instruction is received by the cpu from some software to read memory at 000000324, the next instruction is received from different software to read the same, in a linear world there should be no race.

I don't get it.

What am I missing?

EDIT: If I'm wrong about the cores part, please just consider a single core processor.

 

Edited March 2, 2016 by JohnOne

[TheSaint]

My understanding is that it is to do with pipes, bottlenecks and timing,

Which to me, means you can have simultaneous processes occurring (cores), but then they share output and input pathways, and that is where timing comes into it. First come first served is where I imagine the racing occurs.

I am however far from being an expert, and so could be completely wrong in what my mind has pieced together, from things I have read and what seems logical. It has also been some time between readings ... usually only when considering my next PC.

I have forgotten for instance, where the Southbridge fits into all that ... just remember it improves things in that regard ... alternate pathways or something.

My advice, is don't get old ... you forget or misremember things ..... oops too late.

[jaeger52]

JohnOne, I was intrigued by your question as I realized i did not know the answer either, and I id a little poking around. I think TheSaint is correct about the correlation between race conditions and bottlenecking. It's definitely a timing issue, and bottlenecking could cause enough chaos to have threads run in the wrong sequence, throwing the error. 

[trancexx]

Race condition on level of cpu doesn't exist in "normal" mode. It can happen with overclocking when components that form logic circuits fail to change state fast enough.

JohnOne, I believe you mean operating systems, rather than computers in general. That's where racing conditions can occur, because that's where (artificial) time slicing starts.
It's level above CPU.

[RTFC]

This might clarify.

[jaberwacky]

It seems to me that you would be interested in the dining philosophers problem.

[TheSaint]

Ha Ha, good one Jabbers.

Don't you just love logic puzzles.

[JohnOne]

Great stuff ladies and gentlemen.

So given all the info above, is it safe to say to a layman that race conditions occur because, despite a cpu making linear computations, the speed at which a bit is read cannot be guranteed, perhaps due to the path an electron might take to reach its destination, which is why timing is important?

[JohnOne]

Yes thanks trancexx I should have written OS or software, it is what I meant.

[kcvinu]

  dining philosophers   ..  I am hearing this first time. Read that article. Thanks @jaberwacky  

 

[trancexx]

9 hours ago, JohnOne said:

Great stuff ladies and gentlemen.

So given all the info above, is it safe to say to a layman that race conditions occur because, despite a cpu making linear computations, the speed at which a bit is read cannot be guranteed, perhaps due to the path an electron might take to reach its destination, which is why timing is important?

But that's not it.

When you start Windows you actually start a program (I'll underline it from now on). That program is the only thing that ever "touches" your computer's CPU. Then that program creates interface for you (the user) to start application(s). At the same time the program creates time slots in which it then puts tasks. Time slot is a portion of overall time. The program doesn't advertise how big one time-slot is. It can be two ticks, 1 millisecond, 6 instructions, or whatnot. This is done to create impression that it runs more things at the same time. It actually runs only one thing at any given time, but switches too fast for you to notice it.
When mentioned applications are "started" the program puts it into one time slot. Sometimes these slots are divided into sub-slots (working threads) of a process.
When one time-slot is "active" the program takes binary data (opcodes) from the application and "pushes" it to the CPU onto its behalf.
Race condition can occur, for example, if the program has bug and allows two slots to overlap. This usually ends up with BSOD.
More common and usual case of race condition occurring is when programmer asks the program to create sub-slots (threads) and then access the same addresses from two different sub-slots. Because the program didn't say for how long one sub-slot will run, it can happen that one sub-slot runs for longer time than another before execution is switched to another sub-slot. First slot might have executed more instructions. The problem for the code in another sub-slot is that the code from the first one may have changed data at the shared addresses. Then that other thread continues, but has no idea that the data it works on has been changed by someone else.

Now, that is it in layman's terms.

[jaeger52]

Great rundown, trancexx. That makes a lot more sense to me now!  

[Guest]

I created a very basic example of the race condition bug with my _SharedVar   udf

 

 

The blue window also changes to 100.
But the recorder never catch it (probably for a reason related to the bug I guess? )

Edited March 7, 2016 by Guest

[JohnOne]

Still not proper getting it.

If the cpu only does one thing at a time , which I think is established, then the program, no matter how it might want to cannot tell the cpu to do two things at once.

I'm talking about 2 threads of user process here, each wanting to change value at an address, thread 1 says "change it now" and thread 2 says the same, to me the cpu processes the request of thread 1 and changes value, then it processes request of thread 2 and changes it again, because it can only do one thing at a time, right?

Unless the simple request of thread to change value of lets say an int, is actually broken down by OS into a number of requests, like changing state of just one bit at that address, before switching to the broken down request of thread 2 and attempting to change state of 1 bit for it, and so on until all bits of memory block are changed.

Is that is what is going on?

[trancexx]

18 hours ago, JohnOne said:

Still not proper getting it.

If the cpu only does one thing at a time , which I think is established, then the program, no matter how it might want to cannot tell the cpu to do two things at once.

I'm talking about 2 threads of user process here, each wanting to change value at an address, thread 1 says "change it now" and thread 2 says the same, to me the cpu processes the request of thread 1 and changes value, then it processes request of thread 2 and changes it again, because it can only do one thing at a time, right?

Unless the simple request of thread to change value of lets say an int, is actually broken down by OS into a number of requests, like changing state of just one bit at that address, before switching to the broken down request of thread 2 and attempting to change state of 1 bit for it, and so on until all bits of memory block are changed.

Is that is what is going on?

Imagine two functions running in different threads. Both have access to variable named x.
First function:

x = 1
MakeSureXisLessThan5()
if x > 5 then DestroyTheWorld()
DoNotDestroyTheWorld()

Second one:

Something()
if SomethingElse() then x = 7
if x < 10 then HaveSexOfYourLife()
x = 1

What if during running first thread two lines are executed, then execution is switched to second thread. In this thread also two lines of code is executed and function SomethingElse() returned "true". Then execution in this second thread is paused and switched to first thread. Now, even if in that first thread you called function MakeSureXisLessThan5() to be sure x has correct value when third line is executed the value of x will be 7 and the world will be destroyed.
If in the other thread 4 lines of code have had been executed, the world would still be existing and you would feel much better because of possibly executed line 3.

[czardas]

What would happen if you ran that in reverse? 

Edited March 6, 2016 by czardas

[JohnOne]

See, now I would not have considered that as a race condition, I would have thought that is what should happen, but if ran in a loop would that code crash it's process?

[Bowmore]

I found this short Wikipedia article helped me understand some of what is being discussed in this thread.

https://en.wikipedia.org/wiki/Lamport%27s_bakery_algorithm

However it appears that what I thought of as a race condition previously is something else. I had thought that a race condition occurred when the part of the operating system responsible for allocating slices of time on the CPU was receiving so many request for a slice of time the the process of allocating time slices was consuming all or nearly all the available time slices on the CPU , Much like a Deninal Of Service web attack works. 

[trancexx]

7 hours ago, JohnOne said:

See, now I would not have considered that as a race condition, I would have thought that is what should happen, but if ran in a loop would that code crash it's process?

It would destroy the world. Doesn't that count?

Based on that simple model any kind of code can be written, even such that could crash the process.

[JohnOne]

It would be enough if you wanted to destro the world, but if your process crashed before it could, you'd be pretty pissed off.

I'm still trying to connect the dots here.

When the OS instructs the CPU to change the value of an int, does the CPU carry out that instruction in one go / cycle / or whatever it's called?