Slashdot Mirror
Java Faster Than C++?
jg21 writes "The Java platform has a stigma of being a poor performer, but these new performance benchmark tests suggest otherwise. CS major Keith Lea took time out from his studies at student at Rensselaer Polytechnic Institute in upstate New York's Tech Valley to take the benchmark code for C++ and Java from Doug Bagley's now outdated (Fall 2001) "Great Computer Language Shootout" and run the tests himself. His conclusions include 'no one should ever run the client JVM when given the choice,' and 'Java is significantly faster than optimized C++ in many cases.' Very enterprising performance benchmarking work. Lea is planning next on updating the benchmarks with VC++ compiler on Windows, with JDK 1.5 beta, and might also test with Intel C++ Compiler. This is all great - the more people who know about present-day Java performance, the better.""
I looked at his results page quite extensively, but failed to find a good analysis/justification of the results. Just saying that the Server JVM is better than the Client JVM is *not* enough.
I want to know where the C++ overhead comes from, which Java manages to avoid - does the JVM do better optimization because it is given a better intermediate code (bytecode)? Is it better at doing back/front end optimizations (unlikely given gcc's maturity).
I tried to look for possible discrepancies in the results, but the analysis will definitely take more time - and I think it's the job of the experimenter to do a proper analysis of the results. Liked his choice of benchmarks though.
An Indian-American Hindu committed to non-violent thought/speech/action alarmed by the global explosion of radical Islam
I care that Java is an inconvenient pain to develop in and use. I care that I have to start a mini-OS just to run a Java program. I care that the language is under the control of one vendor. I care that the 'intialization == resource allocation' model doesn't work in Java. I care that the type system is too anemic to support some of the more powerful generic programming constructs. I care that I don't get a choice about garbage collection. I care that I don't get to fiddle bits in particular memory locations, even if I want to.
I think Java is highly overrated. I would prefer that a better C++ (a C-like memory model, powerful generic programming, inheritance, and polymorphism) that lacked C++'s current nightmare of strangely interacting features and syntax.
I use Python when I don't need C++s speed or low-level memory model, and I'm happier for it. It's more flexible than Java, much quicker to develop in, and faster for running small programs. Java doesn't play well with others, and it was seemingly designed not to.
Besides, I suspect that someone who knew and like C++ really well could tweak his benchmarks to make C++ come out faster again anyway. That's something I've noticed about several benchmarks that compare languages in various ways.
Need a Python, C++, Unix, Linux develop
First, it's been known for awhile that Java is a poor performer when writing to the console, for whatever reason. Second, your Java timing probably include the time to startup the VM (not that this is wrong).
If you have a program that runs for awhile (so the startup time is small compared to the time the program takes to run), and does not do intensive output to the console, then Java is a reasonable choice in my opinion. Combined with SWT, Java applications can be quite snappy (see Eclipse, Azureus), and the end user will probably never know the difference.
- shadowmatter
Now, regarding java performance ... Java isn't slow per se, JVMs and some apis (most notably swing) are. Furthermore, JVMs usually have a slow startup, which gave java a bad name (for desktop apps startup matters a lot, for servers it's hardly a big deal). Java can be interpreted, but it doesn't have to be so (all "modern" JVMs compile to binary code on the fly)
Bytecode-based environments will, IMNSHO, eventually lead to faster execution than with pre-compilation. The reason is profiling and specialized code generation. With today's processors, profiling can lead sometimes to spectacular improvements - as much as 30% performance improvements on Itanium for instance. Although Itanium is arguably dead, other future architectures will likely rely on profiling as well. If you don't believe me, check the research in processor architecture and compiling.
The big issue with profiling is that the developper has to do it, and on a dataset that's not necessarily similar to the user's input data. Bytecode environments can do this on-the-fly, and with very accurate data.
The Raven
I'd be interested in comparing the speed of the native code generated by gjc to the that of JVM.
-josh
I am starting on a new standalone server now doing something different, but I am going to stick with Java, and will be happy to see what 1.5 does for me.
But I have seen Java run slow before, and I will tell you this: in every instance it is due to someone writing some needlessly complicated J2EE application with layer upon bloaty layer of indirection. All the wishing in the world won't make one of those behemoths run fast, but it's not fair to blame Java. Maybe blame Sun for EJB's and their best practices, or blame BEA for selling such a pig.
Stuff I like in the Java world:
Here is my experience with C++ vs. Java: At my company, we had a specialized image viewing program. The original program was written in C++ years ago, and performance sucked even on modern machines. It probably had a dozen man-years of time in it. We decided to re-write it in java.
We knew java in theory should be worse than C++ at manipulating large blocks of raw data, so we spent some time architecting, prototyping, and profiling java. We quickly learned the limitations and strengths.
The result? After 4 engineers worked for 6 months, we had a program that was rock solid, had more features, had a modern UI, and was WAY faster. Night and day; the old program felt like a hog, and the new program was zippy as anything. And the new code is fewer lines, and (in our opinion) way more maintainable. Since the original release, we've added severeal new features after day or two of work; the same features never would have happened on the old version, because they would have been too risky.
So the question is this? Could we have re-written or refactored the C++ program and gotten the same speed benefits? No doubt, such a thing is possible. But we are all convinced there is NO WAY we could have done it with as little effort. The C++ version would have taken longer to write and debug.
So, if the JIT computes Hot/Cold Paths, and optimizes the Hot paths, then it should work better and better on successive runs (as more and more profiling information is gathered). On the other hand, there will be cases where it performs worse, as profiles are gathered for specific inputs.
That means that if an average of say 5 runs (on the same input) is taken, it will have an unfair advantage (since gcc did NOT have the advantage of profiling information (see man gprof or similar)). Using Profiling as an optimization tool is *always* unfair unless both tools are provided with the advantage of the same profiling information. This is a valid question for the author then: if the JIT/javac/JVM uses profiling information, gcc should too, for fair comparison.
PS: I have seen this argument being made by my Professor and audiences at compiler conferences.
An Indian-American Hindu committed to non-violent thought/speech/action alarmed by the global explosion of radical Islam
From comments Doug Bagley made about the "Shootout" (where the benchmarks came from), no, I don't think it is an accurate comparison, or at least a conclusive comparison between C++ and Java. His comments from his conclusion:
I put it on the web because I thought it would interest others. Even though I put disclaimers on the page, and I try not to make any claims, I see some people say the shootout shows that "language X is faster than language Y".
That claim is probably premature and hence, bogus. I suppose you could make the claim that, in "Doug's word frequency test, on a PII-450 running Linux 2.4, given a certain input, language X is faster than language Y" Assuming, of course, that I haven't made any mistakes. Some of my tests are also arguably poorly designed and meaningless. (Hey, if you have some better ideas, please write to me).
Benchmarks are notoriously misleading, and perhaps mine aren't any better, although I do try. Benchmarks tell you about results in a very specific case. Drawing a general conclusion is problematic.
One X-factor is JVM warm up. When benchmarking Java you should run the test multiple times in the same VM. This gives you a better real-world feel of what a Java app will do during continuous use, at least from a server perspective.
Desktop app use cases may be different, in which case your test may be valid. Start-up time is definitely a significant part of the user experience. At one point Java 1.5 was supposed to have shared VMs, so that Java can start at system load time. Other VMs would just then be a matter of forking another process off the already running VM, thus increasing startup time. My understanding is this has fallen off the truck for that release, but people are working on it.
======
In X-Windows the client serves YOU!
This didn't exactly fill me with optimism either:
This would seem to imply that the author does not know much about either shell scripting or Makefiles. I'm not sure I'm willing to trust benchmarks from somebody who can't figure out an automated way to build and run them."My life's work has been to prompt others... and be forgotten." --Cyrano de Bergerac
I'm inclined to agree with you, except that the benchmarks were qualified as talking about being relevant to enterprise applications. In such a situation, run-time optimizations are critical.
While it is entirely possible [in c/c++] to use a profiler to generate compiler hints so as to generate even more efficient code, this is rarely performed, and often is not free. A VM otoh does get this capability for free.
Additionally, the java memory manager has a slight edge over tradditional malloc's for total throughput (though the best throughput configurations have horrible spuradic response times). It is also possible to choose a different memory manager for c/c++, but this too is rarely used.. Moreover, it is much harder to have 3'rd party code integrate well with a garbage-collector model. Java enforces garbage collection, and thus optionally gets the particular performance gains (being free to trade off throughput for responsiveness no matter what 3'rd part code is integrated).
As was pointed out, one of the strenghts of C/C++ are pass-by-value, which allows memory allocations to be avoided all-together, but at the cost of copy-time and robustness of code. If a method call requires instantiation, c/c++ have the option of passing in a local [stack resident] structure to be populated by the method. However, this is fodder for buffer-overflow exploits, and notorious for otherwise bad code (accidently caching the address of a value that lives on the stack). Thus, given that c++ will use "new" and thus generally perform a malloc, the same performance issues above apply, and c/c++ may have the additional overhead of copy-by-value.
The fact that you have to explicity declare a c++ parameter as pass-by-reference suggests that those interested in "good programming practices" (tm) will only make a pass-by-reference if you intend to modify it's contents. Thus "clean" code in c++ will be copy-intensive... For fairness, clean java code should always make immutable wrappers for any non-modifyable code, thus requiring an all together different liability (and thus I can't make any claims as to which would be faster; wrapper object instantiation or deep parameter-copy). Though all primatives are available in java as immutable objects (Strings, Dates, etc). Moreoever, clean OO-code should always use method getters, and make all fields private (not even protected). Both C++ and jit'd java can inline these getters.
I haven't looked at the benchmark code, but the above are common components which make a big difference when scaling to large enterprise applications, or even when merely writing a glue application which integrates many large 3'rd party libraries. In c++ you don't have a lot of control over the 3'rd party libraries (in terms of their design trade-offs), but with a VM, you are largely sheltered and have many configurable alternatives.
-Michael
No, it doesn't. Check out WordPerfect Java, Novell ConsoleOne, or any other large Java project for a real world counter example. Java applications are slow to load for any meaningfull piece of client side software. Java works wonderfully for middleware applications but is simply the wrong tool for client side software. When I can reboot the computer and load MCC faster than I can start ConsolOne there is something seriously amiss (and no jokes about having to reboot, I have windows PC's with 200 day uptimes limited only by patching sessions, which is true for any properly maintained OS).
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
I've been playing with those benchmarks for ages. I use them any time a new language comes out or if I just want to do some independent testing.
A couple points:
- The "Great Shootout" benchmark times are sometimes way off because the run-time was too short to get an accurate reading. In those cases the tests should have been run with higher values to really stress the machine. That doesn't appear to be an issue in this test though (assuming his graph values are in seconds).
- Many of the C++ tests are not optimized. That is, they use C++ features like the iostream stuff (cout, and friends) which is extremely slow. The C versions are available and very fast. C++ is pretty much just an extension of C. You don't need to use C++ features if they slow you down. Another one is the hash stuff. In the C++ hash benchmark there are some goofy mistakes made by using the brackets [] operator where it forces several unnecessary lookups. You can also substitute a better STL hashing function that is faster (like MLton's insanely fast hasher).
- The test could be done by comparing C to Java. Anything in C++ can be made as fast as an equivalent C version but there are not many programmers that know how. Just assume anything in C++ will run as fast as a C version, and if it doesn't then you did something wrong. The hash tests would be easier in C++ though. If they were written properly they would kill the Java version.
With that said, I'm going to try these tests myself because I do not believe the results to be accurate. but who knows...
The ratio of people to cake is too big
Also, a quote from the article:
"I was sick of hearing people say Java was slow, when I know it's pretty fast"
Nice, unbiased viewpoint there...
Reviewing the console log, we find that when java programs were tested with a large number of iterations, Java only performed better on one test.
I know that Java has many strengths, but speed isn't one of them. Looking at the results, we see the g++ runtimes are much more consistent than those of Java - on some tests, the Java Server is faster than the client by a factor of 20!? How could a programmer code without having any realistic expectation of the speed of his code. How embarrassed would you be to find that your "blazingly fast" app ran slower than molasses on the client machine, for reasons yet unknown?
When it comes to speed, compiled languages will always run faster than interpreted ones, especially in real-world applications.
But discussions of language speed are a moot point. What this really tested was the implementation, not the language. Speed is never a criteria upon which languages are judged - a "slow" language can always be brought up to speed with compiler optimizations (with a few exceptions). I suspect that if C++ was interpreted, and Java compiled, we'd see exactly the opposite results.
In short, the value of a language consists not in how fast it runs, but in what it enables the programmer to do.
The society for a thought-free internet welcomes you.
Why does the example use a recursive fibonnaci sequence algorithm? It's so slow, and the runtime is dominated by the function call time.
./fib_recurse 40
./fib_for_loop 40
For example:
[bdr@arthurdent tests]$ time
165580141
real 0m3.709s
user 0m3.608s
sys 0m0.005s
time
165580141
real 0m0.006s
user 0m0.002s
sys 0m0.002s
I think a lot of these benchmarks are showing that the Hotspot optimiser is very good at avoiding function call overheads.
Non-graphical Java code can indeed be very competitive with other languages, but it would help if the author bothered to implement the code for his tests intelligently.
The Fibonacci code is recursive, which is about the slowest possible way to implement it, and much of the other code uses high-level features of C++ which are a convenience for the programmer, but are not used when worried about speed.
This fibo code, for example, should be faster:This code was turned in by a student in a lab of mine. This was his first semester in CS, and this code outperforms the Java code quoted on the website considerably. (Try it!).
I am not saying that recursion and high-level C++ features should NOT be used, but I AM saying that if you are comparing the potential speed of languages, you should use tricks that each language provides to optimize speed.
Java will never be faster than properly optimized C++ compiled with an intelligent optimizing compiler except in bizarre corner cases, and tests like this are not terribly convincing demonstrations otherwise. Even the corner cases are removed by a sufficiently talented programmer.
This is also not to say that Java is bad. I think Java is a great language (except for GUI programming with SWING), and definitely makes many programming tasks faster to code and easier to debug than one can do in C++.
Computer Science is no more about computers than astronomy is about telescopes. --E. W. Dijkstra
First of all, the C++ was crappy as many people pointed out.
Second of all, I'm sure that loading the C++ program takes some time more than just loading the byte codes (though that's probably mitigated somewhat by the byte code translation).
Third, the optimization options he used for gcc are a joke. -march=i686 is not even relevant to much larger platforms that can benefit from other optimizations.
And, 4th, and this is the big one, this guy does not know how to benchmark. Anyone who has actually benchmarked their own application knows that if you want to figure out how fast something is, you have to time it IN THE PROGRAM!!!! This would avoid allocation/cout/unnecessary function overhead, when all you're trying to test is a specific operation. I BET (and at some point I will test this) that if you used timing mechanisms INSIDE the programs, that C++ would come out much faster, with the exception of object management and memory stuff (excepting garbage collecting...). Even then, much of that stuff can be overcome by memory pooling, which a surprising number of people ignore.
Until someone does something like all these language comparisons are totally pointless because you are NOT ACTUALLY BENCHMARKING the topic you are looking at. Please lets have someone be intelligent about this for once....
I'm futzing around with the other hash benchmark, and sure enough, making only a trivial change to the code (eliminating the unnecessary strdup in the second hash lookup), gets me about a 30% improvement in performnace.
This guy is a tool.
Let's try not to let fact interfere with our speculation here, OK?
It does mean something. However, here's something a little odd about the differences between C++ and Java: in C++, the compiler makes the decision about what to inline before it fully knows what is calling what. That is because the compiler runs before the linker. So C++ is working with more limited information. Meanwhile, the Java virtual machine has everything integrated. The compiler (the JIT) and the linker (the class loader) and the loader (also part of the class loader) run all together and can cooperate and communicate.
Why is this important? Because the C++ compiler must make all kinds of guesses when it makes these optimizations. If you have a 1-line function, it probably makes sense to inline it in every place it's called, because the speed will increase and the code size won't (the inlined code will probably be smaller than the code to make a function call). But say you have a 25-line function. Should you inline this? If it's called only in one place and nowhere else, it's still more efficient. But if it's called in 50 places, you waste a lot of space by making 25 copies of the function. So maybe you should inline it only in the one or two places where it counts. OK great, but which two places are these? The C++ compiler is left having to just guess. The Java virtual machine can instead just observe the program as it runs and *know* where inlining is worth it and where it isn't.
Also, what if you have a really small function in a library and you want to inline that? Well, in C++, if you dynamically link to that library, you *can't* inline the function. It's just not possible. But with Java, you can, because it's all just classes that are loaded by the class loader and then translated into machine code as needed. So you can inline functions that come from dynamic dependencies, like system libraries.
And then, as someone else said, Java can even inline virtual method calls! How can it do this? Again, by observing the conditions the code is *actually* executing under, not by theorizing about all possible conditions it might execute under. C++ has to allow for the possibility that every object is a different class and so must use a v-table. But Java can, in theory at least, know that while the object *could* be an instance of class X, Y, or Z, in reality the class loader has only loaded class Z, so therefore all instances must be instances of class Z. Presto, inlining is possible, at least until class X or class Y is loaded. But then, since the code is generated dynamically, when you load class X or Y, you could trash the inline code you generated and start over with virtual method calls now that you are having to plan for a different situation.
I should point out that you can get 99% of these performance boosts in C++ by hand-tweaking your code. You can figure out which methods need to not be declared virtual and remove the "virtual" keyword from those methods. If you need to make an inline call to some function in a system library, you can manually copy the code from that library into your source code and call it, as long as you're willing to rewrite and rebuild your C++ app when the library code (that you copied) changes. So these things are theoretically possible, but they are such a code maintenance and system administration nightmare, that they are virtually always far from practical. But with Java or other similar languages, they can all happen for you automatically behind the scenes.
I don't deny that the finer granularity of C++, no make that plain C, no make that raw assembly language, allows you to make certain optimizations that are quite valuable in certain circumstances.
But when you are trying to choose the right tool for a particular job, you need to be current on the details of just what advantages, and what degree of advantage, and in what circumstances, comparing current versions of all of the candidate tools.
This benchmark seems to be showing that things don't just remain the same. That, in fact, the circumstances in which C++ is a better choice than Java are becoming fewer and the advantage in those circumstances is becoming less.
The fact that Java VMs are primarily written in C or C++ indicates that at the time they were initially written, it was believed (I think correctly) that the C or C++ at that time would be a better platform for writing JVMs than the Java of that time, and that since then it has been considered better to extend the existing code than to scrap it and do a complete rewrite in Java.
That's all that this argument proves. Nothing more.
What I'm saying, though, should not be interpreted as a belief that today's Java would be a better choice than today's C++ for writing a Java JVM. I don't know what the relative advantages would be today.
But if C++ were STILL the best tool for writing a JVM from scratch today (certainly possible), that wouldn't mean much when trying to choose a tool for your own app, because most apps bear very little resemblance to a JVM.
"Those who have never entered upon scientific pursuits know not a tithe of the poetry by which they are surrounded."
I've read many times that it actually does ONLY result in a hint to collect. Unless you can prove otherwise, I'm apt to lean in that direction. Which, actually means that the 1:1 implementation is a more realisitic apples to apples comparison. Can I prove that? No. I'm still hoping a java guru will come in with some insightful tidbits. ;)
Write a java app that does nothing but repeatedly call System.gc(). Run it with the -verbose:gc option, and watch the garbage collector go. Mind you, this is not 100% proof, but the fact that it prints out [Full GC] over and over again makes me lean pretty strongly in the direction of "the garbage collector actually runs in response to a System.gc(), it's not just a hint".
One thing I would like Java to do is to allow me to delete objects manually. There are times when the garbage collector really sucks, but 95% of the time it's sufficient, in my experience. And yes, this experience comes from real-world apps.
Regarding the original topic, I would bet that there are cases where Java really could give C++ a run for its money. However, one liability that Java has compared to C at least is that making everything an object adds a whole lot of object overhead. I had to write a file search routine as part of a Java app, and originally wrote it strictly in Java. The sheer number of File objects that get created by such a routine is ridiculous, and there's really no way to reduce the overhead by reusing the objects -- File objects are immutable. Calling out to JNI resulted in a 3 to 5x performance boost. Does this one example prove anything? No, but it's a heck of a lot more real-world than simply appending a string to itself a few dozen times...