Agreed, yet another language war isn't useful. I've written plenty of code for super computing environments in which C/C++ was the primary choice. In that environment I did exactly what you seem to be describing, I altered my coding style to fit the hardware. Teams that are writing low latency tolerant systems (trading) in Java also do the same. They alter their coding style to fit the JVM/Hardware platform they are deploying to. Although we've had more than a decade of rhetoric admonishing the practice, I personally see nothing wrong with that. In those cases the resulting systems are very competitive vs. their C++ counter parts. This isn't to say that Java should be used for all problems. Startup time is a problem for example so short running applications...
As for micro benchmarks, I've seem some really bad stuff done in the name of micro benchmarking and so I'm not sure that Mr. Babbage's great quote applies. He lived in a much different world than we do today ;-) Regards, Kirk On Jul 30, 2010, at 1:46 AM, Alan Kent wrote: > On 29/07/2010 6:19 PM, Kirk wrote: >> Sorry, I just don't buy the standard line that C/C++ is a safe from a >> performance perspective. Array access in Java will be every bit as fast. >> Range checking will most likely be jit'ed out of the code. Direct access >> will most likely be jit'ed into the code.... >> >> Regards, >> Kirk >> > > I can only base it on one personal mid-sized program that I have done myself > (involving lots of XML parsing in Java and C++) which went around 10 times > faster in C++ (a year or two back), and articles I can find doing Google > searches. Anyone who has concrete info would be great. But it really needs > the same program implemented in Java and C++ to do direct comparisons. Some > of the articles have so much subjective sounding text in them I simply do not > believe them. (One more realisitic one involved was I think Quake being > ported to Java and performing just as well - but I don't know if most of the > time is really spent in the GPU rather than Java itself.) > > Based on personal experience (10+ years of C++ programming and 5+ years of > Java programming), the discussions I have had writing performant C++ code > involve talking about memory alignment, cache lines, avoiding memory copies, > templating, inlining, looking at the resultant assembler etc. You can write > multiple classes in C++, then use them to build up a more complex data > structure where the whole structure takes a single malloc to create. Java if > you use multiple classes you get multiple memory allocations (one per object > instance). In C++ I can write an array of classes (or structs) and all the > objects are inlined in the array - in Java I have to have an array of > references to objects, with a new object for each value in the array. In > writing Java you don't have the sort of control as in C++. However in Java > memory management is cheaper (if you have the same number of mallocs!). In > one large (multi-million line C++ code) multi-threaded program, we found > changing the memory allocation library had a 20% difference (or more) in > overall performance. No other change the C++ code - just link in a different > malloc library and major difference in performance. We have had to worry > about things like which threads data structures were allocated from as the > malloc library had a pool per thread. If a different thread ends up doing > the frees, you end up with lots more lock contention in the malloc library. > > I don't want to get carried away here (its all been said before and I did not > mean to start Yet Another Language War), but I have read on the web numerous > opinions (not much evidence) saying Java can generate code around the same > performance as C++ code. I have never seen anything I trust saying it can be > much faster. I have heard (and experienced) cases where its definitely much > slower. I have heard many people I trust in different forums all say C++ > code executes faster - use it when you want to control performance. Its > backed up by personal experience. I have not heard of significant sized > projects where Java saved the day over C++ in terms of performance (that are > backed by believable evidence). What I do believe is Java is much more > productive for programmers, pretty good in performance, and does have harder > to measure benefits in the more modern garbage collectors that only come up > when you have a large running system. Talk to our sysadm admins about Java > and they want to know if anything else needs to run on the same box as if you > have a few Java processes memory consumption goes through the roof (compared > to equivalent C/C++ programs) - making it harder to share a box without > problems. Not trying to be argumentative here, but I have not seen any > evidence that can change my mental model of Java = easier to write and > maintain, C++ = higher performance. (I guess I should add C# = Microsoft.) > > Final word in micro benchmarks, I know their limitiations (have done lots of > performance analysis over the years), but in the words of Charles Babbage > http://en.wikipedia.org/wiki/Charles_Babbage "Errors using inadequate data > are much less than those using no data at all". > > Thanks all! > Alan > > -- > You received this message because you are subscribed to the Google Groups > "The Java Posse" group. > To post to this group, send email to [email protected]. > To unsubscribe from this group, send email to > [email protected]. > For more options, visit this group at > http://groups.google.com/group/javaposse?hl=en. > -- You received this message because you are subscribed to the Google Groups "The Java Posse" group. To post to this group, send email to [email protected]. To unsubscribe from this group, send email to [email protected]. For more options, visit this group at http://groups.google.com/group/javaposse?hl=en.
