Alg. MKL Threaded DGEMV - multithreading

As we all may know, there are lots of different ways to implement DGEMV in parallel (column or block -wise etc) resulting in different communication overheads. I have been looking through both the MKL and all the reference manuals to BLAS to try and figure out which style is in general being called in by cblas_dgemv from MKL(v.11) without success. If anyone has a reference that documents which algorithm or the overheads for the algorithm that is being used, I would be very happy.

MKL ref manuals keep DGEMV as well as other routines as black boxes.
But I think there is still some way to estimate the overhead/efficiency.
As we know, DGEMV is a mem bandwidth bounded operation.
For y += A*x you could measure its speed by the mem bandwidth achieved:
measure the running time for one DGEMV call as t;
compute total mem read/write size: m = 2*len(y)+len(x)+len(A);
actual bandwidth bw = m/t;
check out the peak bandwidth of the total system RAM bw0;
Then bw/bw0*100% can be seen as the actual efficiency of the algorithm.
Please note you may want a large enough matrix/vector to do the measurement. Also if you want repeat the measurement to get more accurate result, you may need to keep the cache cold before starting a new iteration.

Related

Low 'Average Physical Core Utilization' according to VTune when using OpenMP, not sure what the bigger picture is

I have been optimizing a ray tracer, and to get a nice speed up, I used OpenMP generally like follows (C++):
Accelerator accelerator; // Has the data to make tracing way faster
Rays rays; // Makes the rays so they're ready to go
#pragma omp parallel for
for (int y = 0; y < window->height; y++) {
for (int x = 0; x < window->width; x++) {
Ray& ray = rays.get(x, y);
accelerator.trace(ray);
}
}
I gained 4.85x performance on a 6 core/12 thread CPU. I thought I'd get more than that, maybe something like 6-8x... especially when this eats up >= 99% of the processing time of the application.
I want to find out where my performance bottleneck is, so I opened VTune and profiled. Note that I am new to profiling, so maybe this is normal but this is the graph I got:
In particular, this is the 2nd biggest time consumer:
where the 58% is the microarchitecture usage.
Trying to solve this on my own, I went looking for information on this, but the most I could find was on Intel's VTune wiki pages:
Average Physical Core Utilization
Metric Description
The metric shows average physical cores utilization by computations of the application. Spin and Overhead time are not counted. Ideal average CPU utilization is equal to the number of physical CPU cores.
I'm not sure what this is trying to tell me, which leads me to my question:
Is this normal for a result like this? Or is something going wrong somewhere? Is it okay to only see a 4.8x speedup (compared to a theoretical max of 12.0) for something that is embarrassingly parallel? While ray tracing itself can be unfriendly due to the rays bouncing everywhere, I have done what I can to compact the memory and be as cache friendly as possible, use libraries that utilize SIMD for calculations, done countless implementations from the literature to speed things up, and avoided branching as much as possible and do no recursion. I also parallelized the rays so that there's no false sharing AFAIK, since each row is done by one thread so there shouldn't be any cache line writing for any threads (especially since ray traversal is all const). Also the framebuffer is row major, so I was hoping false sharing wouldn't be an issue from that.
I do not know if a profiler will pick up the main loop that is threaded with OpenMP and this is an expected result, or if I have some kind of newbie mistake and I'm not getting the throughput that I want. I also checked that it spawns 12 threads, and OpenMP does.
I guess tl;dr, am I screwing up using OpenMP? From what I gathered, the average physical core utilization is supposed to be up near the average logical core utilization, but I almost certainly have no idea what I'm talking about.
Imho you're doing it right and you overestimate the efficiency of parallel execution. You did not give details about the architecture you're using (CPU, memory etc), nor the code... but to say it simple I suppose that beyond 4.8x speed increase you're hitting the memory bandwidth limit, so RAM speed is your bottleneck.
Why?
As you said, ray tracing is not hard to run in parallel and you're doing it right, so if the CPU is not 100% busy my guess is your memory controller is.
Supposing you're tracing a model (triangles? voxels?) that is in RAM, your rays need to read bits of model when checking for hits. You should check your maximum RAM bandwith, then divide it by 12 (threads) then divide it by the number of rays per second... and find that even 40 GB/s are "not so much" when you trace a lot of rays. That's why GPUs are a better option for ray tracing.
Long story short, I suggest you try to profile memory usage.

Linux: CPU benchmark requiring longer time and different CPU utilization levels

For my research I need a CPU benchmark to do some experiments on my Ubuntu laptop (Ubuntu 15.10, Memory 7.7 GiB, Intel Core i7-4500U CPU # 1.80HGz x 4, 64bit). In an ideal world, I would like to have a benchmark satisfying the following:
The CPU should be an official benchmark rather than created by my own for transparency purposes.
The time needed to execute the benchmark on my laptop should be at least 5 minutes (the more the better).
The benchmark should result in different levels of CPU throughout execution. For example, I don't want a benchmark which permanently keeps the CPU utilization level at around 100% - so I want a benchmark which will make the CPU utilization vary over time.
Especially points 2 and 3 are really key for my research. However, I couldn't find any suitable benchmarks so far. Benchmarks I found so far include: sysbench, CPU Fibonacci, CPU Blowfish, CPU Cryptofish, CPU N-Queens. However, all of them just need a couple of seconds to complete and the utilization level on my laptop is at 100% constantly.
Question: Does anyone know about a suitable benchmark for me? I am also happy to hear any other comments/questions you have. Thank you!
To choose a benchmark, you need to know exactly what you're trying to measure. Your question doesn't include that, so there's not much anyone can tell you without taking a wild guess.
If you're trying to measure how well Turbo clock speed works to make a power-limited CPU like your laptop run faster for bursty workloads (e.g. to compare Haswell against Skylake's new and improved power management), you could just run something trivial that's 1 second on, 2 seconds off, and count how many loop iterations it manages.
The duty cycle and cycle length should be benchmark parameters, so you can make plots. e.g. with very fast on/off cycles, Skylake's faster-reacting Turbo will ramp up faster and drop down to min power faster (leaving more headroom in the bank for the next burst).
The speaker in that talk (the lead architect for power management on Intel CPUs) says that Javascript benchmarks are actually bursty enough for Skylake's power management to give a measurable speedup, unlike most other benchmarks which just peg the CPU at 100% the whole time. So maybe have a look at Javascript benchmarks, if you want to use well-known off-the-shelf benchmarks.
If rolling your own, put a loop-carried dependency chain in the loop, preferably with something that's not too variable in latency across microarchitectures. A long chain of integer adds would work, and Fibonacci is a good way to stop the compiler from optimizing it away. Either pick a fixed iteration count that works well for current CPU speeds, or check the clock every 10M iterations.
Or set a timer that will fire after some time, and have it set a flag that you check inside the loop. (e.g. from a signal handler). Specifically, alarm(2) may be a good choice. Record how many iterations you did in this burst of work.

Is there a standard macro capacity measurement of server compute power?

I'm trying to find a metric for measuring aggregate server capacity. This is not intended to be a direct measure to see if application x, y or z will run on a given machine. It is intended to be used for measuring and comparing compute capacity for large scale workloads across various public cloud vendors and / or private clouds. It is more of an economic measurement than an engineering tool.
My belief is that it would be a measure of cpu capacity without regard to memory or storage.
A rudimentary example might look something like:
(CPU clock speed) * (# of CPUs) * (CPU cores)
This number would most likely be expressed as hertz or some other measure of electricity but again I am going by my rudimentary example above.
This is usually done by comparing benchmark results. naturally each benchmark is evaluating the CPU in a different way, so several different benchmarks are preferable, for example single core, multi-core, etc.
Since each benchmark has a different scale it might be difficult to aggretgate into a single score. That is a question I'd ask in the stats or math stackexchange communities.

Where is the point at which adding additional cores or CPUs doesn’t improve the performance at all?

*Adding a second core or CPU might increase the performance of your parallel program, but it is unlikely to double it. Likewise, a
four-core machine is not going to execute your parallel program four
times as quickly— in part because of the overhead and coordination
described in the previous sections. However, the design of the
computer hardware also limits its ability to scale. You can expect a
significant improvement in performance, but it won’t be 100 percent
per additional core, and there will almost certainly be a point at
which adding additional cores or CPUs doesn’t improve the performance
at all.
*
I read the paragraph above from a book. But I don't get the last sentence.
So, Where is the point at which adding additional cores or CPUs doesn’t improve the performance at all?
If you take a serial program and a parallel version of the same program then the parallel program has to do some operations that the serial program does not, specifically operations concerned with coordinating the operations of the multiple processors. These contribute to what is often called 'parallel overhead' -- additional work that a parallel program has to do. This is one of the factors that makes it difficult to get 2x speed-up on 2 processors, 4x on 4 or 32000x on 32000 processors.
If you examine the code of a parallel program you will often find segments which are serial, that is which only use one processor while the others are idle. There are some (fragments of) algorithms which are not parallelisable, and there are some operations which are often not parallelised but which could be: I/O operations for instance, to parallelise these you need some sort of parallel I/O system. This 'serial fraction' provides an irreducible minimum time for your computation. Amdahl's Law explains this, and that article provides a useful starting point for your further reading.
Even when you do have a program which is well parallelised the scaling (ie the way speed-up changes as the number of processors increases) does not equal 1. For most parallel programs the size of the parallel overhead (or the amount of processor time which is devoted to operations which are only necessary for parallel computing) increases as some function of the number of processors. This often means that adding processors adds parallel overhead and at some point in the scaling of your program and jobs the increase in overhead cancels out (or even reverses) the increase in processor power. The article on Amdahl's Law also covers Gustafson's Law which is relevant here.
I've phrased this all in very general terms, no consideration of current processor and computer architectures; what I am describing are features of parallel computation (as currently understood) not of any particular program or computer.
I flat out disagree with #Daniel Pittman's assertion that these issues are of only theoretical concern. Some of us are working very hard to make our programs scale to very large numbers of processors (1000s). And almost all desktop and office development these days, and most mobile development too, targets multi-processor systems and using all those cores is a major concern.
Finally, to answer your question, at what point does adding processors no longer increase execution speed, now that is an architecture- and program-dependent question. Happily, it is one that is amenable to empirical investigation. Figuring out the scalability of parallel programs, and identifying ways of improving it, are a growing niche within the software engineering 'profession'.
#High Performance Mark is right. This happens when you are trying to solve a fixed size problem in the fastest possible way, so that Amdahl' law applies. It does not (usually) happen when you are trying to solve in a fixed time a problem. In the former case, you are willing to use the same amount of time to solve a problem
whose size is bigger;
whose size is exactly the same as before, but with a greeter accuracy.
In this situation, Gustafson's law applies.
So, let's go back to fixed size problems.
In the speedup formula you can distinguish these components:
Inherently sequential computations: σ(n)
Potentially parallel computations: ϕ(n)
Overhead (Communication operations etc): κ(n,p)
and the speedup for p processors for a problem size n is
Adding processors reduces the computation time but increases the communication time (for message-passing algorithms; it increases the synchronization overhead etcfor shared-memory algorithm); if we continue adding more processors, at some point the communication time increase will be larger than the corresponding computation time decrease.
When this happens, the parallel execution time begins to increase.
Speedup is inversely proportional to execution time, so that its curve begins to decline.
For any fixed problem size, there is an optimum number of processors that minimizes the overall parallel execution time.
Here is how you can compute exactly (analytical solution in closed form) the point at which you get no benefit by adding additional processors (or cores if you prefer).
The answer is, of course, "it depends", but in the current world of shared memory multi-processors the short version is "when traffic coordinating shared memory or other resources consumes all available bus bandwidth and/or CPU time".
That is a very theoretical problem, though. Almost nothing scales well enough to keep taking advantage of more cores at small numbers. Few applications benefit from 4, less from 8, and almost none from 64 cores today - well below any theoretical limitations on performance.
If we're talking x86 that architecture is more or less at its limits. # 3 GHz electricity travels 10 cm (actually somewhat less) per Hz, the die is about 1 cm square, components have to be able to switch states in that single Hz (1/3000000000 of a second). The current manufacturing process (22nm) gives interconnections that are 88 (silicon) atoms wide (I may have misunderstood this). With this in mind you realize that there isn't that much more that can be done with physics here (how narrow can an interconnection be? 10 atoms? 20?). At the other end the manufacturer, to be able to market a device as "higher performing" than its predecessor, adds a core which theoretically doubles the processing power.
"Theoretically" is not actually completely true. Some specially written applications will subdivide a large problem into parts that are small enough to be contained inside a single core and its exclusive caches (L1 & L2). A part is given to the core and it processes for a significant amount of time without accessing the L3 cache or RAM (which it shares with other cores and therefore will be where collisions/bottlenecks will occur). Upon completion it writes its results to RAM and receives a new part of the problem to work on.
If a core spends 99% of its time doing internal processing and 1% reading from and writing to shared memory (L3 cache and RAM) you could have an additional 99 cores doing the same thing because, in the end, the limiting factor will be the number of accesses the shared memory is capable of. Given my example of 99:1 such an application could make efficient use of 100 cores.
With more common programs - office, ie, etc - the extra processing power available will hardly be noticed. Some parts of the programs may have smaller parts written to take advantage of multiple cores and if you know which ones you may notice that those parts of the programs are much faster.
The 3 GHz was used as an example because it works well with the speed of light which is 300000000 meters/sec. I read recently that AMD's latest architecture was able to execute at 5 GHz but this was with special coolers and, even then, it was slower (processed less) than an intel i7 running at a significantly slower frequency.
It heavily depends on your program architecture/design. Adding cores improves parallel processing. If your program is not doing anything in parallel but only sequentially, adding cores would not improve its performance at all. It might improve other things though like framework internal processing (if you're using a framework).
So the more parallel processing is allowed in your program the better it scales with more cores. But if your program has limits on parallel processing (by design or nature of data) it will not scale indefinitely. It takes a lot of effort to make program run on hundreds of cores mainly because of growing overhead, resource locking and required data coordination. The most powerful supercomputers are indeed massively multi-core but writing programs that can utilize them is a significant effort and they can only show their power in an inherently parallel tasks.

Severe multi-threaded memory bottleneck after reaching a specific number of cores

We are testing our software for the first time on a machine with > 12 cores for scalability and we are encountering a nasty drop in performance after the 12th thread is added. After spending a couple days on this, we are stumped regarding what to try next.
The test system is a dual Opteron 6174 (2x12 cores) with 16 GB of memory, Windows Server 2008 R2.
Basically, performance peaks from 10 - 12 threads, then drops off a cliff and is soon performing work at about the same rate it was with about 4 threads. The drop-off is fairly steep and by 16 - 20 threads it reaches bottom in terms of throughput. We have tested both with a single process running multiple threads and as multiple processes running single threads-- the results are pretty much the same. The processing is fairly memory intensive and somewhat disk intensive.
We are fairly certain this is a memory bottleneck, but we don't believe it a cache issue. The evidence is as follows:
CPU usages continues to climb from 50 to 100% when scaling from 12 to 24 threads. If we were having synchronization/deadlock issues, we would have expected CPU usage to top out before reaching 100%.
Testing while copying a large amount of files in the background had very little impact on the processing rates. We think this rules out disk i/o as the bottleneck.
The commit charge is only about 4 GBs, so we should be well below the threshold in which paging would become an issue.
The best data comes from using AMD's CodeAnalyst tool. CodeAnalyst shows the windows kernel goes from taking about 6% of the cpu time with 12 threads to 80-90% of CPU time when using 24 threads. A vast majority of that time is spent in the ExAcquireResourceSharedLite (50%) and KeAcquireInStackQueuedSpinLockAtDpcLevel (46%) functions. Here are the highlights of the kernel's factor change when going from running with 12 threads to running with 24:
Instructions: 5.56 (times more)
Clock cycles: 10.39
Memory operations: 4.58
Cache miss ratio: 0.25 (actual cache miss ratio is 0.1, 4 times smaller than with 12 threads)
Avg cache miss latency: 8.92
Total cache miss latency: 6.69
Mem bank load conflict: 11.32
Mem bank store conflict: 2.73
Mem forwarded: 7.42
We thought this might be evidence of the problem described in this paper, however we found that pinning each worker thread/process to a particular core didn't improve the results at all (if anything, performance got a little worse).
So that's where we're at. Any ideas on the precise cause of this bottleneck or how we might avoid it?
I'm not sure that I understand the issues completely such that I can offer you a solution but from what you've explained I may have some alternative view points which may be of help.
I program in C so what works for me may not be applicable in your case.
Your processors have 12MB of L3 and 6MB of L2 which is big but in my view they're seldom big enough!
You're probably using rdtsc for timing individual sections. When I use it I have a statistics structure into which I send the measurement results from different parts of the executing code. Average, minimum, maximum and number of observations are obvious but also standard deviation has its place in that it can help you decide whether a large maximum value should be researched or not. Standard deviation only needs to be calculated when it needs to be read out: until then it can be stored in its components (n, sum x, sum x^2). Unless you're timing very short sequences you can omit the preceding synchronizing instruction. Make sure you quantifiy the timing overhead, if only to be able to rule it out as insignificant.
When I program multi-threaded I try to make each core's/thread's task as "memory limited" as possible. By memory limited I mean not doing things which requires unnecessary memory access. Unnecessary memory access usually means as much inline code as possible and as litte OS access as possible. To me the OS is a great unknown in terms of how much memory work a call to it will generate so I try to keep calls to it to a minimum. In the same manner but usually to a lesser performance impacting extent I try to avoid calling application functions: if they must be called I'd rather they didn't call a lot of other stuff.
In the same manner I minimize memory allocations: if I need several I add them together into one and then subdivide that one big allocation into smaller ones. This will help later allocations in that they will need to loop through fewer blocks before finding the block returned. I only block initialize when absolutely necessary.
I also try to reduce code size by inlining. When moving/setting small blocks of memory I prefer using intrinsics based on rep movsb and rep stosb rather than calling memcopy/memset which are usually both optimized for larger blocks and not especially limited in size.
I've only recently begun using spinlocks but I implement them such that they become inline (anything is better than calling the OS!). I guess the OS alternative is critical sections and though they are fast local spinlocks are faster. Since they perform additional processing it means that they prevent application processing from being performed during that time. This is the implementation:
inline void spinlock_init (SPINLOCK *slp)
{
slp->lock_part=0;
}
inline char spinlock_failed (SPINLOCK *slp)
{
return (char) __xchg (&slp->lock_part,1);
}
Or more elaborate (but not overly so):
inline char spinlock_failed (SPINLOCK *slp)
{
if (__xchg (&slp->lock_part,1)==1) return 1;
slp->count_part=1;
return 0;
}
And to release
inline void spinlock_leave (SPINLOCK *slp)
{
slp->lock_part=0;
}
Or
inline void spinlock_leave (SPINLOCK *slp)
{
if (slp->count_part==0) __breakpoint ();
if (--slp->count_part==0) slp->lock_part=0;
}
The count part is something I've brought along from embedded (and other programming) where it is used for handling nested interrupts.
I'm also a big fan of IOCPs for their efficiency in handling IO events and threads but your description does not indicate whether your application could use them. In any case you appear to economize on them, which is good.
To address your bullet points:
1) If you have 12 cores at 100% usage and 12 cores idle, then your total CPU usage would be 50%. If your synchronization is spinlock-esque, then your threads would still be saturating their CPUs even while not accomplishing useful work.
2) skipped
3) I agree with your conclusion. In the future, you should know that Perfmon has a counter: Process\Page Faults/sec that can verify this.
4) If you don't have the private symbols for ntoskrnl, CodeAnalyst may not be able to tell you the correct function names in its profile. Rather, it can only point to the nearest function for which it has symbols. Can you get stack traces with the profiles using CodeAnalyst? This could help you determine what operation your threads perform that drives the kernel usage.
Also, my former team at Microsoft has provided a number of tools and guidelines for performance analysis here, including taking stack traces on CPU profiles.

Resources