My app is written in Java.
There is a C++ library I need to utilize. I don't want to use JNI.
60 times a second, the C++ app needs to send the Java app 10MB of data; and the Java app needs to send the C++ app 10 MB of data.
Both apps are running on the same machine; the OS is either Linux or Mac OS X.
What is the most efficient way to do this? (At the moment, I'm considering TCPIP ports; but in C++, I can do memory mapping -- can I do something similar in Java?)
Thanks!
Yes, Java has memory-mapped files with the NIO framework.
If you're trying to avoid JNI because you didn't want to write stubs, you can also interface with C++ code (at least ones that are extern "C") using JNA. For best performance, use direct mapping (concrete classes with native methods, not a mapped interface)---see documentation for more details. :-)
Using mapped files is a way of hand-rolling a highly optimized rpc. You might consider starting with a web service talking over local sockets, using MTOM for attaching the data, or just dropping it into a file. Then you could measure the performance. If the data was a problem, you could then use mapping.
Note that there are some odd restrictions on this that make your code sensitive to whether it is running on Windows or not. On Windows, you can't delete something that is open.
I should point out that I have done exactly what you are proposing here. It has a control channel on a socket, and the data is shared via a file that is mmapped in C++ (or the Windows equivalent) and NIO mapped in Java. It works. I've never measured maximum throughput, though.
Sounds like shared memory would be the way to go. I believe the NIO libraries support that.
The question is WHAT do you want your program to do?
You should give a look at BridJ (and JNAerator).
It's a recent alternative to JNA with support for C++ and a special focus on performance.
Not directly helpful and will be interesting at least to develop but you could throw in an SSD/RAM drive accessible to both the Java and C++ application and have a sort of juggle between data ops in there with file-based locking and all that odd stuff.
What would make this scheme sort of manageable from performance point is that for this purpose Java NIO has ByteBuffer which is a high level representation of low level byte mapping on disk.
You should take a look at javolution.
They are using NIO direct buffers for data exchange.
In theory, this should be faster then plain JNI
since you do not have the overhead of passing
data through the argument list.
Related
Is it possible to write an antivirus program in Java such as that it can intercept a program from being executed? Can I have such a deep control of the OS in Java?
update:
what about c#? same restrictions apply or that is a better way?
Having such influence on the OS is possible. There is only the problem, that you will lose the platform independency or at least have to write the code for every given platform due to the reason that such actions require quite deep access of the system which could be achived with JNI, which would tie the method you use it in to the OS.
I don't think that sort of control is possible with Java, primarily because it uses a VM and is shielded from the OS. Or rather the OS is shielded from the Java VM. This is by design.
Edited to add for clarity: I am assuming that you want to write the entire solution in Java, and not mix languages.
I am not convinced that it would work even with JNI.
In the case of "intercepting" when the OS starts a new process (or writes to a file or whatever), you need to write some kind of driver or kernel module which hooks into the OS. That driver/module is most certainly written in native compiled code. So the OS is the one in charge here, and will eventually call your native module.
So, as I see it, Java is not even involved here.
Thats the basic approach anyway. It may be possible using something like pam in Linux which is configurable to do almost anything related to security and file/process permissions and can call other processes to do its bidding. Seems far fetched though to run a JVM instance for each new process the OS tries to start.
As HalloDu said, this is technically possible with the use of JNI. However, IIRC, most antivirus programs use some sort of driver to intercept opened files and scan them before allowing the OS to continue using the file. This being the case, the amount of native code you would have to write (in C or possibly C++) would be substantial and is likely to outstrip your Java code in size.
When writing low-ish level apps, I'd stick to C. However, it might make sense to code things like the GUI in a higher level language, though Java wouldn't be my choice there either, because it's kind of a pain to interface with C. Personally, I'd do the whole damn thing in C just because mixing languages tends to be a pain. If I had to mix languages, my choices would by C and python, simply because ctypes makes interfacing with C really easy.
It is possible with the JNI. You would mostly be using Java for a GUI and C/C++ for any other sort of antivirus work though.
What is the point in making your own Antivirus? It is a lot of work, but I guess it would be cool if you made it a portable one that block and removes all the more nasty ones. If you must persist, ClamAV, it is an open source and pretty good AV (no realtime protection) but programmed in C++.
Your best bet might be to write the GUI and much of the logic in Java, then have a C or C++ back-end that does the scans.
You can then re-use the front-end across platforms and keep the platform specific stuff in the lower levels.
This way you can use the strengths of both languages--Java's platform independence and ease of use and C/C++'s ability to directly access the underlying platform.
I want to send the raw audio buffer to c++ for audio transcoding.
I have two option
using piped stream
using direct buffers (java.nio)
Are these really my 2 best options (and which would people recommend?) Thanks!
JNI is easy to screw up, therefore people tend to complain about it. But it is an excellent and extremely stable option, when done correctly.
Direct buffers in NIO will almost certainly have better performance. This is pretty much the ideal case for direct buffers.
I'm not sure what the point of your question is - if you want to know if there are other options, then the answer is certainly yet (you could, for example, write to a file then invoke an external application to process it - or you could use JNI without direct buffers). But if you want a tightly coupled, highly performant interface between Java and C++ code, JNI along with direct buffers, is going to be the tool to use.
You may use sockets (that accepts connection from local) but that will be more effective on *nix systems I guess.
If you don't need realtime transcoding, I would choose the buffers approach. In this way you have better control and you'll be sure that no byte is lost.
I know of at least one post which has same words like this. But this is not exactly same as that post. I'm trying to work a way to "share" data between a .NET and Java application. I'm not concerned about objects, but just plain strings if u like.
I have a .NET application capturing real-time data and a Java application which has capability to analyze and work on this data. I'm looking for ways to re-use this same java app without coding it entirely in .NET.
My problem is that the data is "fairly" REAL-Time (.NET), and so has to be the analysis (Java). I can live with microsecond delays but I can't afford one second delay. WebServices, Queues (as in Messaging Queues), RDBMS are some of the options I can think of. Is there any better way?
Or has anybody got some real performance numbers for the solutions I mentioned above to select one of them? And just to get started: RDBMSs' are not "THAT" good for concurrent (connections doing) insertion/updation/reading, at least with the crude way of doing DBMS stuff. (Deadlocks?)
What are "objects" if not a mechanism for describing "data"? But I digress - I suspect I would look at a TCP socket between the two. If the data is very basic, then fine - just write directly to the stream; if there is any complexity, perhaps use something like "protocol buffers" to provide an easy way of reading/writing dense data to a stream without having to write every last byte yourself.
I think microsecond delays are going to be a challenge for any approach here... will millisecond delays do?
For completeness:
Another possible is to use Named pipes, it should be pretty quick, and I'd imagine (being a java guy I can only imagine) that .NET has native support for them. The down side is that on windows you'll have to either write a JNI extension or use a library like JNA to poke around at the Win32 API from Java.
Sounds like a local socket could do. The latency should be in low ms or less.
Depending on your program you may get some milage out of what #Cowan reports in answer to 'Any Concept of shared memory in java', his answer is: Any concept of shared memory in Java
In summary: he say's that you can use memory mapped files between two processes on the same machine. This in theory could work between .NET and java assuming .NET has some memory mapped file support.
Different machines communicate with each other by sending messages into sockets. Please check the below link for example.
Socket programming in the real world
Answers provided here are great. One idea that might be of interest, but is probably asking for more trouble than it's worth is to load both VMs in a single process (both the JVM and the CLR can be loaded within a native Windows application) and give them access to native code. Java via JNI and .Net via the mapping functions to native code that they allow.
You could also leverage native queue semaphores to wake up a thread on one side or the other when data is updated.
While JNI transitions are expense, they would probably still be faster than the native local socket implementation.
How is your Java application currently deployed? It sounds to me like you're willing to make some modification to it, so I'm assuming you have access to the source code.
I know this is a little out there, but could you compile the Java application in the J# compiler, so that your .NET app has native access to it?
You can convert your compiled java application to .NET by IKVM. After that you can change logic of your .NET application so it will not make data transfers to Java application, but just call data processing code written in Java as it were written and compiled for .NET.
There are a number of JMS servers which support .NET and Java clients. These can perform messages in under a millisecond.
However you might like to try an RPC solution like Hessian RPC or Protobuf RPC. These can achieve lower latencies and can give the appearance of direct calls between platforms. These support .NET and Java as well.
I have a really simple Java class that effectively decorates a Map with input validation, with the obvious void set() and String get() methods.
I'd like to be able to effectively call those methods and handle return values and exceptions from outside the JVM, but still on the same machine Update: the caller I have in mind is not another JVM; thanks #Dave Ray
My implementation considerations are typical
performance
ease of implementation and maintenance (simplicity?)
reliability
flexibility (i.e. can I call from a remote machine, etc.)
Is there a 'right way?' If not, what are my options, and what are the pro/cons for each?
(Stuff people have actually done and can provide real-life feedback on would be great!)
Ok. Here's another try now that I know the client is not Java. Since you want out-of-process access and possibly remote machine access, I don't think JNI is what you want since that's strictly in-process (and a total hassle). Here are some other options:
Raw Sockets : just set up a listener socket in Java and accept connections. When you get a connection read the request and send back a response. Almost every language can use sockets so this is a pretty universal solution. However, you'll have to define your own marshalling scheme, parsing, etc.
XML-RPC : this isn't as hip these days, but it's simple and effective. There are Java libraries as well as libraries in most other languages.
CORBA : as mentioned above, CORBA is an option, but it's pretty complicated and experts are getting harder to come by.
Web Server : set up an embedded web server in your app and handle reqests. I've heard good things about Jetty or you can use the one provided with Java. I've used the latter successfully to server KML files to Google Earth from a simulation written in Java. Most other languages have libraries for making HTTP requests. How you encode the data (XML, text, etc) is up to you.
Web Services : This would be more complicated I think, but you could use JAX-WS to expose you objects as web services. NetBeans has pretty nice tools for building Web Services, but this may be overkill.
Will you be calling from another JVM-based system, or is the client language arbitrary? If you're calling from another JVM, one of the simplest approaches is to expose your object as an MBean through JMX. The canonical Hello World MBean is shown here. The pros are:
Really easy to implement
Really easy to call from other JVMs
Support for remote machines
jconsole allows you to manually test your MBean without writing a client
Cons:
Client has to be on a JVM (I think)
Not great for more complicated data structures and interactions. For example, I don't think an MBean can return a reference to another MBean. It will serialize and return a copy.
Since your callers are not Java apps and you're already foreseeing networked callers, RMI-IIOP (CORBA) might be an option. Though it's definitely not easy to implement, it has the advantage of being a widely-recognized standard.
Since your caller is not JVM-based, this is a question of inter-process communication with JVM. The options I have in mind are:
Communicate over a socket: make your JVM listen to incoming connections and caller send commands
Communicate using shared files (caller writes to file, JVM polls and updates)
Using JNI, start JVM inside a callers process and then use RMI/MBeans to communicate with the first ("server") JVM. Caller will have access to results using JNI
Option 3 IMO is the most "Java" way of doing this, and is the most complex/error-prone.
Option 2 is ugly but simple
Option 1 is moderately easy (java part) and otherwise ok.
For ease of use, I would use Spring Remoting. If you are already using Spring in your project, that's a no brainer. If you arent ... well you should have a look anyway.
Spring provides an abstraction that allow you to switch remoting protocols easily. It supports the most widely deployed protocols (SOAP, Hessian, Burlap, RMI, ...). If you are calling from non Java code, Hessian has support in a number of other languages, is known to be more efficient than SOAP and easier than CORBA.
Beanshell is a shell-like java interpreter that can be exposed over a network socket. Basically you do this from java:
i = new bsh.Interpreter();
i.set( "myapp", this ); // Provide a reference to your app
i.eval("server(7000)");
and then you do this from anywhere else:
telnet localhost 7001
myapp.someMethod();
This little utility does remote java invocations much more easily than JNI or RMI ever has.
For more, start at: http://www.beanshell.org/manual/remotemode.html
JNI (Java Native Interface) allows access to java code from C or C++.
I have an Inno Setup script (installing a Java program) which calls some Java methods to perform some operations or check some conditions.
I (actually my predecessor) just instanciate java.exe on each call. Which is, obviously, costly, although not critical in my case (and the Windows cache kicks in, I suppose).
An alternative is to use some inter-language communication/messaging, your Java program acting as a server. Corba comes to mind, as it is language agnostic. But a bit heavy, perhaps. You can use sockets. RPC is another buzzword too, but I haven't much experience in the field.
What you want is the Java Native Interface (JNI), despite the difficulties that it may present. There is no other equivalent technology that will be as easy to implement.
As mentioned in the comments for the preceding answer, the JNI is optimized for calling native code from Java, but it can also be used for the reverse with a little work. In your native code you'll need to implement the JNI entry point--something like SetMapPointer()--then call that function from the Java code once the Map is built. The implementation of SetMapPointer() should save the Java object pointer someplace accessible, then the native code can invoke Java methods on it as needed.
You'll need to make sure that this happens in the right order (i.e. the native code doesn't try to access the Map before it's been built and passed to native code), but that shouldn't be an especially hard problem.
Another alternative to consider if the other process will be on the same machine and the OS is POSIX-compliant (not Windows) is Named Pipes.
The outside process writes the operations, as strings or some other agreed-upon byte encoding, to the named pipe while the Java application is reading from the pipe, parsing up the incoming operations and executing them against your object.
This is the same strategy that you would use for socket connections, just instead of a SocketInputStream you'd be reading from a FileInputStream that is attached to a named pipe.
An alternative to CORBA is ICE, unless the licence is a problem (it's GPL, but you can also buy a commercial licence).
It has pretty much all the benefits of CORBA, but ZeroC, the vendor, provides bindings for many different languages. CORBA vendors tend to only provide one or two language bindings, and then you start finding compatibility problems.
The documentation is also excellent. I wouldn't have said it was particularly easy to pick up, but probably easier than CORBA.
Otherwise, another option I don't think has been mentioned is the new middleware/RPC framework developed by Cisco, now donated to Apache, called Etch. It's still pretty new though, and documentation is sparse.
I've been reading Skype4Java (java api for skype) and noticed they use jni to access the skype client.
intuitively I'd assume that there already is a standard library in java that has an OS-sensitive jni implementation to access other processes. I set up to look for one, but couldn't find it.
Is there such a library? if not, is there a best practice to access another process in the os, not necessarily a skype client?
From Java 1.4 onwards you can use memory mapped files to exchange arbitrary information with another process. See java.nio.MappedByteBuffer for details.
I think that maybe you need to define what 'access' means to you. IF you are talking about plain old inter-process communication, then sockets or JNI are really your best bet.
Garth's comment about using memory mapped files is interesting - I've used MMFs and virtual files for IPC between C applications many times, but it never occurred to me that Java's MMF implementation might be compatible with the native OS virtual file system. These kinds of virtual files usually require non-trivial setup, so I'd be surprised if it would work...
All said, unless you are pumping massive amounts of data between apps, using sockets is probably the most universal and effective way of doing it. Be sure you account for endianness between the host OS and Java VM :-)
This is not quite what you're looking for, but will probably help a great deal nonetheless: the Java Native Access project on java.net.
Don't forget sockets...