Related
I have implemented an OpenMP/MPI hybrid parallel algorithm, and would like to measure its strong-scaling parallel efficiency. For this, I would have to calculate speed-up S=t(1)/t(N), and then the efficiency E=S/N.
Background: Having done some analysis, I was able to show that the peak efficiency of the algorithm could be expected at a problem size, at which the single node of my benchmark cluster cannot house the data required.
Possible solutions: I can either:
calculate speed-up using the smallest node-count, at which the data can be housed e.g. at 4 nodes => S=t(4)/t(N), or,
calculate the theoretical single-node time-to-solution t(1) by extrapolation, and then use that value as reference.
Questions:
Which approach is better and why?
If I use the first approach, can I, strictly speaking, even refer to it as strong-scaling parallel efficiency, seeing as it doesn't conform to the definition provided above?
Bonus question: When we measure t(1), should we run the algorithm with simulated communication calls (i.e. by calling mpirun -n 1 ./my_benchmark_program), or should we rather call a version of the program which performs no communication at all (i.e. ./my_openmp_only_benchmark_program)?
I hope this post is clear, please ask for clarification if it isn't. Any help will be greatly appreciated. Thanks in advance.
There are various problems with the classical definition of speedup if you are using MPI. The single processor case involves no communication, while the two-processor one does, so there is overhead in the t(2) case and it will always be less than twice as fast. This is even worse if you have a multicore/multinode setup, where up to 16 (or so) processes will run on a single node, so t(17) will suddenly be much slower because it starts involving a second node.
This means you can not simply apply the textbook formulas. You need to explain how you are doing your scalability study. For instance: one process per node until the number of processes == number of nodes, then start putting multple processes on each node, et cetera.
The fact that the single-process case does not fit in memory is then a minor hiccup: you start with a base case of multiple processes, and document that fact, plus your reasoning for the base case that you actually used.
I've been doing a lot of studying from many different resources on algorithm analysis lately, and one thing I'm currently confused about is why time complexity is often defined in terms of the number of steps/operations an algorithm performs.
For instance, in Introduction to Algorithms, 3rd Edition by Cormen, he states:
The running time of an algorithm on a particular input is the number of primitive operations or “steps” executed. It is convenient to define the notion of step so that it is as machine-independent as possible.
I've seen other resources define the time complexity as such as well. I have a problem with this because, for one, it's called TIME complexity, not "step complexity" or "operations complexity." Secondly, while it's not a definitive source, an answer to a post here on Stackoverflow states "Running time is how long it takes a program to run. Time complexity is a description of the asymptotic behavior of running time as input size tends to infinity." Further, on the Wikipedia page for time complexity it states "In computer science, the time complexity is the computational complexity that describes the amount of computer time it takes to run an algorithm." Again, these are definitive sources, things makes logical sense using these definitions.
When analyzing an algorithm and deriving its time complexity function, such as in Figure 1 below, you get an equation that is in units of time. It CAN represent the amount of operations the algorithm performs, but only if those constant factors (C_1, C_2, C_3, etc.) are each a value of 1.
Figure 1
So with all that said, I'm just wondering how it's possible for this to be defined as the number of steps when that's not really what it represents. I'm trying to clear things up and make the connection between time and number of operations. I feel like there is a lot of information that hasn't been explicitly stated in the resources I've studied. Hoping someone can help clear things up for me, and without going into discussion about Big-O because that shouldn't be needed and misses the point of the question, in my opinion.
Thank you everyone for your time and help.
why time complexity is often defined in terms of the number of steps/operations an algorithm performs?
TL;DR: because that is how the asymptotic analysis work; also, do not forget, that time is a relative thing.
Longer story:
Measuring the performance in time, as we, humans understand the time in a daily use, doesn't make much sense, as it is not always that trivial task to do.. furthermore - it even makes no sense in a broader perspective.
How would you measure what is the space and time your algorithm takes? what will be the conditional and predefined unit of the measurement you're going to apply to see the running time/space complexity of your algorithm?
You can measure it on your clock, or use some libraries/API to see exactly how many seconds/minutes/megabytes your algorithm took.. or etc.
However, this all will be VERY much variable! because, the time/space your algorithm took, will depend on:
Particular hardware you're using (architecture, CPU, RAM, etc.);
Particular programming language;
Operating System;
Compiler, you used to compile your high-level code into lower abstraction;
Other environment-specific details (sometimes, even on the temperature.. as CPUs might be scaling operating frequency dynamically)..
therefore, it is not the good thing to measure your complexity in the precise timing (again, as we understand the timing on this planet).
So, if you want to know the complexity (let's say time complexity) of your algorithm, why would it make sense to have a different time for different machines, OSes, and etc.? Algorithm Complexity Analysis is not about measuring runtime on a particular machine, but about having a clear and mathematically defined precise boundaries for the best, average and worst cases.
I hope this makes sense.
Fine, we finally get to the point, that algorithm analysis should be done as a standalone, mathematical complexity analysis.. which would not care what is the machine, OS, system architecture, or anything else (apart from algorithm itself), as we need to observe the logical abstraction, without caring about whether you're running it on Windows 10, Intel Core2Duo, or Arch Linux, Intel i7, or your mobile phone.
What's left?
Best (so far) way for the algorithm analysis, is to do the Asymptotic Analysis, which is an abstract analysis calculated on the basis of input.. and that is counting almost all the steps and operations performed in the algorithm, proportionally to your input.
This way you can speak about the Algorithm, per se, instead of being dependent on the surrounding circumstances.
Moreover; not only we shouldn't care about machine or peripheral factors, we also shouldn't care about Lower Order Terms and Constant Factors in the mathematical expression of the Asymptotic Analysis.
Constant Factors:
Constant Factors are instructions which are independent from the Input data. i.e. which are NOT dependent on the input argument data.
Few reasons why you should ignore them are:
Different programming language syntaxes, as well as their compiled files, will have different number of constant operations/factors;
Different Hardware will give different run-time for the same constant factors.
So, you should eliminate thinking about analyzing constant factors and overrule/ignore them. Only focus on only input-related important factors; therefore:
O(2n) == O(5n) and all these are O(n);
6n2 == 10n2 and all these are n2.
One more reason why we won't care about constant factors is that they we usually want to measure the complexity for sufficiently large inputs.. and when the input grows to the + infinity, it really makes no sense whether you have n or 2n.
Lower order terms:
Similar concept applies in this point:
Lower order terms, by definition, become increasingly irrelevant as you focus on large inputs.
When you have 5x4+24x2+5, you will never care much on exponent that is less than 4.
Time complexity is not about measuring how long an algorithm takes in terms of seconds. It's about comparing different algorithms, how they will perform with a certain amount if input data. And how this performance develops when the input data gets bigger.
In this context, the "number of steps" is an abstract concept for time, that can be compared independently from any hardware. Ie you can't tell how long it will take to execute 1000 steps, without exact specifications of your hardware (and how long one step will take). But you can always tell, that executing 2000 steps will take about twice as long as executing 1000 steps.
And you can't really discuss time complexity without going into Big-O, because that's what it is.
You should note that Algorithms are more abstract than programs. You check two algorithms on a paper or book and you want to analyze which works faster for an input data of size N. So you must analyze them with logic and statements. You can also run them on a computer and measure the time, but that's not proof.
Moreover, different computers execute programs at different speeds. It depends on CPU speed, RAM, and many other conditions. Even a program on a single computer may be run at different speeds depending on available resources at a time.
So, time for algorithms must be independent of how long a single atomic instruction takes to be executed on a specific computer. It's considered just one step or O(1). Also, we aren't interested in constants. For example, it doesn't matter if a program has two or 10 instructions. Both will be run on a fraction of microseconds. Usually, the number of instructions is limited and they are all run fast on computers. What is important are instructions or loops whose execution depends on a variable, which could be the size of the input to the program.
This is the first time i ask question here so thanks very much in advance and please forgive my ignorance. And also I've just started to CUDA programming.
Basically, i have a bunch of points, and i want to calculate all the pair-wise distances. Currently my kernel function just holds on one point, and iteratively read in all other points (from global memory), and conduct the calculation. Here's some of my confusions:
I'm using a Tesla M2050 with 448 cores. But my current parallel version (kernel<<<128,16,16>>>) achieves a much higher parallelism (about 600x faster than kernel<<<1,1,1>>>). Is it possibly due to the multithreading thing or pipeline issue, or they actually indicate the same thing?
I want to further improve the performance. So i figure to use shared memory to hold some input points for each multiprocessing block. But the new code is just as fast. What's the possible cause? Could it be related to the fact that i set too many threads?
Or, is it because i have a if-statement in the code? The thing is, i only consider and count the short distances, so i have a statement like (if dist < 200). How much should i worry about this one?
A million thanks!
Bin
Mark Harris has a very good presentation about optimizing CUDA: Optimizing Parallel Reduction in CUDA.
Algorithmic optimizations
Changes to addressing, algorithm cascading
11.84x speedup, combined!
Code optimizations
Loop unrolling
2.54x speedup, combined
Having an extra operations statement, does indeed cause problems although it will be the last thing you want to optimize, if not simply because you need to know the layout of your code before implementing the size assumptions!
The problem you are working on sounds like the famous n-body problem,
see Fast N-Body Simulation with CUDA.
An additional performance increase can be achieved if you can avoid doing a pairwise computation, for example, the elements are too far to have an effect on each-other. This applies to any relationship that can be expressed geometrically, whether it be pairwise costs or a physics simulation with springs. My favorite method is to divide the grid into boxes and, with each element putting itself into a box via division, then only evaluate pairwise relations between between neighboring boxes. This can be called O(n*m).
(1) The GPU runs many more threads in parallel than there are cores. This is because each core is pipelined. Operations take around 20 cycles on compute capability 2.0 (Fermi) architectures. So for each clock cycle, the core starts work on a new operation, returns the finished result of one operation, and move all the other (around 18) operations one more step towards completion. So, to saturate the GPU, you might need something like 448 * 20 threads.
(2) It's probably because your values are getting cached in the L1 and L2 caches.
(3) It depends on how much work you're doing inside the if conditional. The GPU must run all 32 threads in a warp through all the code inside the if even if the condition is true for only a single of those threads. If there is a lot of code in the conditional as compared to the rest of your kernel, and relatively view threads go through that code path, it is likely that you end up with low compute throughput.
I have some very heavy code to develop and want to make some calculations beforehand.
Now I'm trying to make a very rough estimate with MIPS, but can not find anything about what MIPS actually stands for. Is an instruction a single bitwise operation/comparison in MIPS?
The best thing you can do is to run your algorithm on some (much) smaller set N. If you can estimate the complexity of your algorithms, you can then estimate how fast it will run for the full dataset.
MIPS is not a good way to go; in most algorithms, CPU spends more than half of time waiting for caches/RAM anyway; only small set of problems allows for very good analysis on how is memory going to be used (e.g. matrix operations) and can be tuned to use CPU efficiently.
I was interested to know about parameters other than space and time during analysing the effectiveness of an algorithms. For example, we can focus on the effective trap function while developing encryption algorithms. What other things can you think of ?
First and foremost there's correctness. Make sure your algorithm always works, no matter what the input. Even for input that the algorithm is not designed to handle, you should print an error mesage, not crash the entire application. If you use greedy algorithms, make sure they truly work in every case, not just a few cases you tried by hand.
Then there's practical efficiency. An O(N2) algorithm can be a lot faster than an O(N) algorithm in practice. Do actual tests and don't rely on theoretical results too much.
Then there's ease of implementation. You usually don't need the best intro sort implementation to sort an array of 100 integers once, so don't bother.
Look for worst cases in your algorithms and if possible, try to avoid them. If you have a generally fast algorithm but with a very bad worst case, consider detecting that worst case and solving it using another algorithm that is generally slower but better for that single case.
Consider space and time tradeoffs. If you can afford the memory in order to get better speeds, there's probably no reason not to do it, especially if you really need the speed. If you can't afford the memory but can afford to be slower, do that.
If you can, use existing libraries. Don't roll your own multiprecision library if you can use GMP for example. For C++, stuff like boost and even the STL containers and algorithms have been worked on for years by an army of people and are most likely better than you can do alone.
Stability (sorting) - Does the algorithm maintain the relative order of equal elements?
Numeric Stability - Is the algorithm prone to error when very large or small real numbers are used?
Correctness - Does the algorithm always give the correct answer? If not, what is the margin of error?
Generality - Does the algorithm work in many situation (e.g. with many different data types)?
Compactness - Is the program for the algorithm concise?
Parallelizability - How well does performance scale when the number of concurrent threads of execution are increased?
Cache Awareness - Is the algorithm designed to maximize use of the computer's cache?
Cache Obliviousness - Is the algorithm tuned for particulary cache-sizes / cache-line-sizes or does it perform well regardless of the parameters of the cache?
Complexity. 2 algorithms being the same in all other respects, the one that's much simpler is going to be a much better candidate for future customization and use.
Ease of parallelization. Depending on your use case, it might not make any difference or, on the other hand, make the algorithm useless because it can't use 10000 cores.
Stability - some algorithms may "blow up" with certain test conditions, e.g. take an inordinately long time to execute, or use an inordinately large amount of memory, or perhaps not even terminate.
For algorithms that perform floating point operations, the accumulation of round-off error is often a consideration.
Power consumption, for embedded algorithms (think smartcards).
One important parameter that is frequently measure in the analysis of algorithms is that of Cache hits and cache misses. While this is a very implementation and architecture dependent issue, it is possible to generalise somewhat. One particularly interesting property of the algorithm is being Cache-oblivious, which means that the algorithm will use the cache optimally on multiple machines with different cache sizes and structures without modification.
Time and space are the big ones, and they seem so plain and definitive, whereby they should often be qualified (1). The fact that the OP uses the word "parameter" rather than say "criteria" or "properties" is somewhat indicative of this (as if a big O value on time and on space was sufficient to frame the underlying algorithm).
Other criteria include:
domain of applicability
complexity
mathematical tractability
definitiveness of outcome
ease of tuning (may be tied to "complexity" and "tactability" afore mentioned)
ability of running the algorithm in a parallel fashion
(1) "qualified": As hinted in other answers, a -technically- O(n^2) algorithm may be found to be faster than say an O(n) algorithm, in 90% of the cases (which, btw, may turn out to be 100% of the practical cases)
worst case and best case are also interesting, especially when linked to some conditions in the input. if your input data shows some properties, an algorithm, by taking advantage of this property, may perform better that another algorithm which performs the same task but does not use that property.
for example, many sorting algorithm perform very efficiently when input are partially ordered in a specific way which minimizes the number of operations the algorithm has to execute.
(if your input is mostly sorted, an insertion sort will fit nicely, while you would never use that algorithm otherwise)
If we're talking about algorithms in general, then (in the real world) you might have to think about CPU/filesystem(read/write operations)/bandwidth usage.
True they are way down there in the list of things you need worry about these days, but given a massive enough volume of data and cheap enough infrastructure you might have to tweak your code to ease up on one or the other.
What you are interested aren’t parameters, rather they are intrinsic properties of an algorithm.
Anyway, another property you might be interested in, and analyse an algorithm for, concerns heuristics (or rather, approximation algorithms), i.e. algorithms which don’t find an exact solution but rather one that is (hopefully) good enough.
You can analyze how far a solution is from the theoretical optimal solution in the worst case. For example, an existing algorithm (forgot which one) approximates the optimal travelling salesman tour by a factor of two, i.e. in the worst case it’s twice as long as the optimal tour.
Another metric concerns randomized algorithms where randomization is used to prevent unwanted worst-case behaviours. One example is randomized quicksort; quicksort has a worst-case running time of O(n2) which we want to avoid. By shuffling the array beforehand we can avoid the worst-case (i.e. an already sorted array) with a very high probability. Just how high this probability is can be important to know; this is another intrinsic property of the algorithm that can be analyzed using stochastic.
For numeric algorithms, there's also the property of continuity: that is, whether if you change input slightly, output also changes only slightly. See also Continuity analysis of programs on Lambda The Ultimate for a discussion and a link to an academical paper.
For lazy languages, there's also strictness: f is called strict if f _|_ = _|_ (where _|_ denotes the bottom (in the sense of domain theory), a computation that can't produce a result due to non-termination, errors etc.), otherwise it is non-strict. For example, the function \x -> 5 is non-strict, because (\x -> 5) _|_ = 5, whereas \x -> x + 1 is strict.
Another property is determinicity: whether the result of the algorithm (or its other properties, such as running time or space consumption) depends solely on its input.
All these things in the other answers about the quality of various algorithms are important and should be considered.
But time and space are two things that vary at some rate compared to the size of the input (n). So what else can vary according to n?
There are several that are related to I/O. For example, the number of writes to a disk is an important one, which may not be directly shown by space and time estimates alone. This becomes particularly important with flash memory, where the number of writes to the same memory location is the significant metric in some algorithms.
Another I/O metric would be "chattiness". A networking protocol might send shorter messages more often adding up to the same space and time as another networking protocol, but some aspect of the system (perhaps billing?) might make minimizing either the size or number of the messages desireable.
And that brings us to Cost, which is a very important algorithmic consideration sometimes. The cost of an algorithm may be affected by both space and time in different amounts (consider the separate costing of server storage space and gigabits of data transfer), but the cost is the thing that you wish to minimize overall, so it may have its own big-O estimations.