You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@harmony.apache.org by "Zakharov, Vasily M" <va...@intel.com> on 2006/04/27 21:13:46 UTC
RE: [rmi] package comparison
Daniel,
Yeah, structurizing is always good. :)
a) You provided an excellent list of spec issues. Sure, I'll reply to it
soon.
b) Well, cross-review could be a good and interesting idea, however I'm
not sure it would give us much significant information to make
decisions. I think we should care about how it works first, then on how
it's built. See d) I'm adding below.
By the way, you could check 'doc' subdirectory in our contribution, it
describes most implementation features that are different from RI.
c) Well, performance is important, but less important than the
functionality, I think. However, measuring performance is always good
anyway. Do you have any ideas of particular applications we could use?
d) The most important thing in an RMI implementation, as far as I can
see, is its capability to functionally support the real existing
applications. I think this is the first and most important thing we
should check. In other words, we have to select a set of real apps
actively using RMI functionality, and check both implementations on
them.
Vasily
-----Original Message-----
From: Daniel Gandara [mailto:danielgandara@neosur.com]
Sent: Monday, April 24, 2006 6:26 PM
To: harmony-dev@incubator.apache.org
Subject: Re: [rmi] package comparison (was Re: Contribution of RMI
framework)
Hi Vasily,
I believe we should sum up and start specific threads for each topic;
This is a list of topics and posting names I suggest:
a) Improve Specification --> [rmi] improve spec
We agreed in the "problems" the spec has. I believe here -Harmony-
is the place to start a request for improvement. I started a post
"[rmi] spec issues" but noone commented on it, I guess I should
repost it :)
b) Packages --> [rmi] <specific issue>
In order to end up with a better package for the community, we
should review each other implementation and architecture approach,
after that, we can discuss on specific things.
c) Performance & real apps --> [rmi] real apps
We need to define a set of real apps and workload to run against
the packages in order to measure performance. I see a similar
discusion taking place at java.math, since they also have to compare
two implementations.
please comment on each, and add/remove if I'm missing something.
Thanks,
Daniel
----- Original Message -----
From: "Zakharov, Vasily M" <va...@intel.com>
To: <ha...@incubator.apache.org>
Sent: Friday, April 21, 2006 12:56 PM
Subject: RE: [rmi] package comparison (was Re: Contribution of RMI
framework)
Daniel,
> black-box testing (reverse engineering I would call the complete
procedure)
Black-box testing is NOT reverse engineering, as the latter could
suggest disassembling the code or something like that, that would
violate the clean-room procedures and that we certainly didn't do. What
we were doing is checking carefully how our implementation works with
reference implementation, and making sure the RI gets what it asks for
in the protocol.
> is an approach, but, can you be sure you are 100% compatible?
No, of course we can't, as we didn't pass the Java TCK. :) But we were
testing interoperability extensively, and also tried the real
applications.
> what about other providers?
I don't know, we didn't check any other providers. But if they're
compatible with RI, they would most probably be also compatible with our
implementation. :)
> so we should now move on comparing what is comparable between the
packages: its functionality.
Yeah, I agree with that.
> I mean, you can provide compatibility with reference implementation
> (let's say provider A), but what about users working with provider B?
> or C? are you assuming all providers do the same black-boxing? and
> if so, all of them arrived to the same conclusion about how the wire
> protocol behaves?
Surely, we can do nothing about this (except improving the spec, which
is surely good, but a very long process). If all providers strive to be
compatible with reference implementation - they probably would be
compatible to each other. If they don't - they would never be
compatible.
Also, the reference implementation is still most widely used, so being
compatible to it is most important.
> Standards and specifications are what interoperability is all about!
Yes, of course, I fully agree with this.
> there is no way in which you can provide compatibility if there is no
standard;
And this is the thing I can't agree with. Standard greatly helps in
achieving compatibility, but that can also be done without a standard.
> so I strongly believe that having a better quality spec is the
> way we can really benefit the community ... if not, you -and I and
> other providers too- will always be one step behind the "reference"
> implementation, spending lots of hours and resources finding
> out how things are being done inside the black box.
Yes, of course, improving the spec is a great goal. But it's a long term
goal, and I completely agree with you on this - in long term. But
improving the spec would probably take years, and we surely need some
short term solution to provide to the community now.
> it will be great if you contribute what you have conclude about the
protocol.
Hmm, we've already done that - it's already in the code. We weren't
writing any documentation about the internal details of our
implementation, we were only noting things in the doc where our
implememtation behaved differently than the RI.
Vasily
-----Original Message-----
From: Daniel Gandara [mailto:danielgandara@neosur.com]
Sent: Friday, April 21, 2006 1:17 AM
To: harmony-dev@incubator.apache.org
Subject: Re: [rmi] package comparison (was Re: Contribution of RMI
framework)
Zakharov, Vasily M wrote
>Daniel,
>
>> We started our development as a clean room implementation
>> of the package following the spec; and we found it -the spec
>> (JRMP included)- to be vague and incomplete making it impossible
>> for us to get interoperability without doing "code inspection"
>> or "reverse engineering", which was not allowed by the imposed
>> clean-room restriction. I believe you faced the same situation,
>> and it will be good to know how you tackle it without compromising
>> your development.
>
>Of course, we also followed the clean-room restrictions, but we
>struggled to be compatible with reference implementation and used
>a lot of black-box testing to find out what goes on in the reference
>implementation data exchange.
black-box testing (reverse engineering I would call the complete
procedure)
is an approach, but, can you be sure you are 100% compatible?
what about other providers?
>> Having said that, I believe comparision between packages should
>> be done based on functionality and NOT on interoperability since
>> it is -at least- underspecified and it can be acomplished with
further
>> wire protocol information.
>
>Regretfully, I can't agree with this approach. Our target is providing
>the end users out there with a free open source Java implementation
>they can really use in their applications.
I completelly agree with you on the ultimate goal; but now I believe we
should compare, comparable things... it is clear that "interoperability"
is a feature you have worked and acomplished, in the other hand we
have worked and acomplished on 5.0; so we should now move on
comparing what is comparable between the packages: its functionality.
>And users generally don't care about quality of specifications
>what they need is the tool they can take and use, and that will work
>and satisfy their needs. That's why I think it's very important to not
let
>the poor quality of the specification stop us from giving the community
>an effective and compatible implementation.
I see two different issues here, so I'll comment each:
a) end users
>" And users generally don't care about quality of specifications
> what they need is the tool they can take and use ..."
That is true, but black-box inspecting method is not good enough,
I mean, you can provide compatibility with reference implementation
(let's say provider A), but what about users working with provider B?
or C? are you assuming all providers do the same black-boxing? and
if so, all of them arrived to the same conclusion about how the wire
protocol behaves?
b) the spec
> "... not let the poor quality of the specification stop us from giving
the
> community an effective and compatible implementation ..."
Standards and specifications are what interoperability is all about!
there is no way in which you can provide compatibility if there is no
standard; so I strongly believe that having a better quality spec is the
way we can really benefit the community ... if not, you -and I and
other providers too- will always be one step behind the "reference"
implementation, spending lots of hours and resources finding
out how things are being done inside the black box. If that be the
case then we should no longer talk about "specification" for rmi,
instead
we should talk about rmi as a "product" we try to emulate...
>> Needeless to say, I'm aware of the critical impact interoperability
>> has and we are currently working on that.
>By the way, one of the very important interoperability issues is the
>1.1 protocol version. It's really obsolete, and there's a great
>temptation to give up on implementing it, but the reference
implementation
> strictly adhers to 1.1 protocol in Distributed Garbage Collection, so
>we must have 1.1 protocol operational in order to be practically
>compatible with reference implementation.
Yes that's true and there is a big challenge to be 100% compatible
supporting different reference implementation's evolutions.
BTW: it will be great if you contribute what you have conclude about
the protocol.
Daniel
>
>Vasily Zakharov
>Intel Middleware Products Division
>
-----Original Message-----
From: Daniel Gandara [mailto:danielgandara@neosur.com]
Sent: Thursday, April 20, 2006 1:17 AM
To: harmony-dev@incubator.apache.org
Subject: Re: [rmi] package comparison (was Re: Contribution of RMI
framework)
Vasily,
You are not missing anything, our package does not allow
interoperability simply because you cannot derive it from the
spec. We started our development as a clean room implementation
of the package following the spec; and we found it -the spec (JRMP
included)- to be vague and incomplete making it impossible for us to
get interoperability without doing "code inspection" or "reverse
engineering", which was not allowed by the imposed clean-room
restriction. I believe you faced the same situation, and it will be
good to know how you tackle it without compromising your development.
Our strategic decision at that moment was to move forward on the
development being sure about the "cleanliness" and left the reverse
engineering of the wire protocol to be done later after the package
is complete. This is the main reasons why I sent the "[rmi] spec
issues"
post presenting the problems we found on the spec; which I believe are
strong enough to ask for a JSR or at least a clarification on the spec.
For a detailed list of the issues we found on the spec please go to
http://www.itc.unc.edu.ar/javadev/rmi/specissues.html .
Having said that, I believe comparision between packages should
be done based on functionality and NOT on interoperability since it
is -at least- underspecified and it can be acomplished with further wire
protocol information. Needeless to say, I'm aware of the critical
impact
interoperability has and we are currently working on that.
Daniel
PS: if you do have further information/description of the JRMP protocol
please send it ot me, since it will be very usefull.
----- Original Message -----
From: "Zakharov, Vasily M" <va...@intel.com>
To: <ha...@incubator.apache.org>
Sent: Tuesday, April 18, 2006 2:57 PM
Subject: RE: [rmi] package comparison (was Re: Contribution of RMI
framework)
Daniel,
I've been trying to do some comparisons, as I promised, and I believe
I'm missing something.
I was testing the interoperability, and when I tried to use an "Intel"
RMI client against an "ITC" server, it failed, although it worked
against a "Sun" server. Changing client and server produces the same
result. Here're the stack traces I get, the test code and the config.
Any idea, what I'm
doing wrong?
Vasily Zakharov
Intel Middleware Products Division
Stack trace 1 ("ITC" client, "Sun" or "Intel" server):
java.rmi.ConnectIOException: I/O exception Creating Connection; nested
exception is:
java.rmi.MarshalException: Exception marshaling JRMP Header;
nested exception is:
java.rmi.UnmarshalException: Exception reading the Header
response; nested exception is:
java.io.EOFException
at
ar.org.fitc.rmi.transport.ConnectionPool.getClientConnection(Unknown
Source)
at ar.org.fitc.rmi.transport.TransportManager.invoke(Unknown
Source)
at ar.org.fitc.rmi.server.UnicastRemoteRefImpl.invoke(Unknown
Source)
at ar.org.fitc.rmi.registry.RegistryImpl_Stub.lookup(Unknown
Source)
at Client.main(Client.java:14)
Caused by: java.rmi.MarshalException: Exception marshaling JRMP Header;
nested exception is:
java.rmi.UnmarshalException: Exception reading the Header
response; nested exception is:
java.io.EOFException
at
ar.org.fitc.rmi.transport.StreamClientConnection.handshake(Unknown
Source)
at
ar.org.fitc.rmi.transport.StreamClientConnection.establishConnection(Unk
nown Source)
... 5 more
Caused by: java.rmi.UnmarshalException: Exception reading the Header
response; nested exception is:
java.io.EOFException
at
ar.org.fitc.rmi.transport.jrmp.ClientProtocolHandler.readHandshakeRespon
se(Unknown Source)
... 7 more
Caused by: java.io.EOFException
at java.io.DataInputStream.readByte(Unknown Source)
... 8 more
Stack trace 2 ("Intel" client, "ITC" server):
java.rmi.ConnectIOException: Unable to acknowledge protocol with server;
nested exception is:
java.io.EOFException
at
org.apache.harmony.rmi.transport.tcp.TcpConnection.serverProtocolAck(Tcp
Connection.java:145)
at
org.apache.harmony.rmi.client.ClientConnection.<init>(ClientConnection.j
ava:90)
at
org.apache.harmony.rmi.transport.tcp.TcpConnection.<init>(TcpConnection.
java:73)
at
org.apache.harmony.rmi.client.ClientConnectionManager.getConnection(Clie
ntConnectionManager.java:107)
at
org.apache.harmony.rmi.remoteref.UnicastRef.newCall(UnicastRef.java:226)
at
org.apache.harmony.rmi.remoteref.UnicastRef.invoke(UnicastRef.java:127)
at
org.apache.harmony.rmi.registry.RegistryImpl_Stub.lookup(RegistryImpl_St
ub.java:134)
at Client.main(Client.java:14)
Caused by: java.io.EOFException
at java.io.DataInputStream.readByte(Unknown Source)
at
org.apache.harmony.rmi.transport.tcp.TcpConnection.serverProtocolAck(Tcp
Connection.java:112)
... 7 more
Server.java:
import java.rmi.Remote;
import java.rmi.RemoteException;
import java.rmi.RMISecurityManager;
import java.rmi.registry.Registry;
import java.rmi.registry.LocateRegistry;
import java.rmi.server.UnicastRemoteObject;
interface TestRemoteInterface extends Remote {
public String test() throws RemoteException;
}
class TestRemoteObject implements TestRemoteInterface {
public String test() throws RemoteException {
System.out.println("TestRemoteObject.test() run");
return "SUCCESS";
}
}
public class Server {
public static void main(String[] args) {
TestRemoteObject obj = null;
try {
System.out.println("Setting security manager");
System.setSecurityManager(new RMISecurityManager());
System.err.println("Creating registry");
Registry registry = LocateRegistry.createRegistry(1099);
System.out.println("Creating remote object");
obj = new TestRemoteObject();
System.out.println("Exporting remote object");
UnicastRemoteObject.exportObject(obj, 5555);
System.out.println("Binding object to registry");
registry.rebind("TestRemoteObject", obj);
System.out.println("Sleeping 10 seconds");
Thread.sleep(10000);
} catch (Throwable e) {
e.printStackTrace();
System.err.println("ERROR");
} finally {
if (obj != null) {
System.out.println("Unexporting remote object");
try {
if (UnicastRemoteObject.unexportObject(obj, false))
{
System.out.println("Unexport FALSE");
} else {
if (UnicastRemoteObject.unexportObject(obj,
true)) {
System.out.println("Unexport TRUE");
} else {
System.out.println("Unexport FAILED");
System.err.println("ERROR");
System.exit(-1);
}
}
} catch (Throwable e) {
e.printStackTrace();
System.err.println("ERROR");
System.exit(-1);
}
}
}
}
}
Client.java:
import java.rmi.RMISecurityManager;
import java.rmi.registry.Registry;
import java.rmi.registry.LocateRegistry;
public class Client {
public static void main(String[] args) {
try {
System.out.println("Setting security manager");
System.setSecurityManager(new RMISecurityManager());
System.err.println("Locating registry");
Registry registry = LocateRegistry.getRegistry();
System.err.println("Looking for object");
TestRemoteInterface obj = (TestRemoteInterface)
registry.lookup("TestRemoteObject");
System.err.println("Object found: " + obj);
System.err.println("Calling method");
Object ret = obj.test();
System.err.println("Object returned: " + ret);
System.err.println("SUCCESS");
} catch (Throwable e) {
e.printStackTrace();
System.err.println("FAIL");
}
}
}
all.policy:
grant{
permission java.security.AllPermission;
};
-----Original Message-----
From: Daniel Gandara [mailto:danielgandara@neosur.com]
Sent: Monday, April 17, 2006 7:38 PM
To: harmony-dev@incubator.apache.org
Subject: [rmi] package comparison (was Re: Contribution of RMI
framework)
Vasily,
a couple of things about package comparison:
a) java 5.0 vs 1.4.2
Our rmi package was developed according to 5.0 rmi spec, and
it takes full advantage of 5.0 new features (like java.util.concurrent)
Since Harmony classlib and VMs are still in 1.4.2 we deployed
a 1.4.2 version of our package in which we removed the 5.0
dependencies. There is obviously a performance penalty
paid by the 1.4.2 release of the package.
You can find both versions of the packages at
http://issues.apache.org/jira/browse/HARMONY-211
b) compatibility with reference implementation
within our contribution you will find a complete set of test cases
(source code and documentation for each). We run these test cases
against our package before contributing it, so I guess one way to
move further is to cross run test cases (you run ours and we run
yours). What do you think?
c) performance analysis and comparison
I believe the first step here is to get along about which is the
workload or set of applications that represent a "real" use of rmi
package. I see a big challenge here...
I'll wait for your comments,
Daniel
----- Original Message -----
From: "Zakharov, Vasily M" <va...@intel.com>
To: <ha...@incubator.apache.org>
Sent: Friday, April 14, 2006 1:17 PM
Subject: RE: Contribution of RMI framework
Hi, Mikhail,
Regretfully, the method-to-method comparison would hardly be effective
with RMI, as it's a highly integrated component.
80% of implementation is hidden in non-public API, and some components
(e. g. RMIC) have no public API at all. So it's hard to plug just one
public method from one implementation to another without modifying
non-public code - and non-public code could have (and probably does
have) very different structure in different implementations.
What we really can do is try to run both these implementations and
compare them for conformance to the specification, compatibility with
reference implementation, maybe stability, performance, visual code
quality etc. I'm now planning to do some of these, so we'd get some
results pretty soon.
Vasily
-----Original Message-----
From: Mikhail Loenko [mailto:mloenko@gmail.com]
Sent: Friday, April 14, 2006 7:53 AM
To: harmony-dev@incubator.apache.org
Subject: Re: Contribution of RMI framework
I think we need compare contributions method by method to assemble
the best classlib
Thanks,
Mikhail
2006/4/14, Daniel Gandara <da...@neosur.com>:
> Vasily,
> good to know that there is someone out there who has also
> been working on rmi; I believe we'll have a lot to share and discuss
> about it.
>
> Thanks,
>
> Daniel
>
> ----- Original Message -----
> From: "Zakharov, Vasily M" <va...@intel.com>
> To: <ha...@incubator.apache.org>
> Sent: Wednesday, April 12, 2006 9:53 PM
> Subject: Contribution of RMI framework
>
>
> Hi, all,
>
> I would like to announce the next code contribution to Harmony project
> on
> behalf of Intel corporation. This contribution contains the
> implementation
> of RMI framework.
>
> The archive with this contribution can be found at:
>
> http://issues.apache.org/jira/browse/HARMONY-337
>
> The Remote Method Invocation (RMI) framework enables an object in one
> virtual machine to call methods of an object in another one, to create
> applications distributed on various Java virtual machines on the same
> or different hosts.
>
> For more information please see the documentation contained in the
> bundle.
>
> The code is a result of efforts of Intel Middleware Product Division
> team.
> One should be able to run this code with a 1.4+ compatible JRE/VM (was
> tested using commercial VMs). No classes require special support from
> the VM.
> All code is pure Java. The implementation is done according to Java
1.4
> specification of RMI.
>
> The archive contains the README file that explains the building and
> running
> process for this code. If any additional comments or clarifications
are
> needed, feel free to contact me. I will be happy to answer all
questions
> about this contribution and to participate in its further
> development/maintenance and integration into Harmony.
>
> Vasily Zakharov
> Intel Middleware Product Division
>
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org
---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org