diff options
Diffstat (limited to 'TAO/docs/releasenotes/orbcore.html')
| -rw-r--r-- | TAO/docs/releasenotes/orbcore.html | 477 |
1 files changed, 0 insertions, 477 deletions
diff --git a/TAO/docs/releasenotes/orbcore.html b/TAO/docs/releasenotes/orbcore.html deleted file mode 100644 index 8ae4cb68dca..00000000000 --- a/TAO/docs/releasenotes/orbcore.html +++ /dev/null @@ -1,477 +0,0 @@ -<!DOCTYPE HTML PUBLIC "-//IETF//DTD HTML//EN"> - -<!-- $Id$ --> - -<html> - <head> - <title>ORB & ORB Core Status</title> - </head> - - <HR> <H3><A NAME="orb"></A>ORB & ORB Core</H3> Point of - contact: <A HREF="mailto:cleeland@cs.wustl.edu">Chris Cleeland</A> - - <p>Last Update: $Date$ </P> - - <P>Current status: - - <UL> - <LI>Provides dynamically linked wholesale assignment of ORB - strategies for communication, concurrency, demultiplexing, - scheduling, and dispatching using the ACE Service Configurator - framework.</LI> - - <LI>Supports several concurrency models out of the box: - Reactive, Thread-per-Connection, and ORB-per-thread. The - ORB-per-thread design is optimized for real-time I/O - subsystems.</LI> - - <LI>Supports <TT>resolve_initial_references("NameService")</TT> - using ACE support for multicast.</LI> - </UL> - - <p>Known issues: - <dl> - <dt><em>Servers cannot properly report OS-chosen port.</em></dt> - - <dd>It is often desireable for a server to start up and not have - to specify the TCP port number on which the ORB will listen, - i.e., let the OS decide. While this works properly (by - providing an argument <code>-ORBport 0</code>), it is very - difficult for the server to publish the IOR for clients' use. - This is because the port is not assigned until - <code>ORB::run</code> is called.</dd> - - <dt><em>Demarshalling problems with complex types</em></dt> - - <dd>The current demarshalling scheme may not work in cases of - complex types such as sequences of unions or anys or object - references. The engine has difficulty determining the dynamic - type of the objects.</dd> - - <dt><em>Object references for objects dynamically created within - a server may not have proper host.</em> (See also <a - href="#hostfix">recent fix</a>.)</dt> - - <dd>When an object is dynamically created and its object - reference returned, it simply has to "guess" at the host - information that is returned in the IIOP Profile. Currently, it - guesses using the host name, which is a reasonable guess in most - cases.<p> - - However, if there are multiple interfaces on a machine - <i>and</i> they have different host names, it is possible that - the desired effect will not be realized, i.e., the connection to - that new object may use a different interface from the original - object. Currently the only way to change this is to use the <a - href="../Options.html#-ORBhost"><b>-ORBhost</b> option</a> to - manually specify that dynamically-created objects bind to a - specific interface so that their IIOP Profile has the desired - hostname. A more desirable fix is being investigated.</dd> - - <dt><em>ORB-per-thread and thread-per-connection models cannot - be mixed.</em></dt> - - <dd> - <p> - These two concurrency models do not currently mix because - there is no way for the connection-handling threads to gain - access to the resources held by the ORB in thread-specific - storage (TSS). In order to do this, the Resource Manager - will have to implement some sort of inheritance of - resources. - <p> - There are other problems which will need to be solved as - well for this to work properly. For example, similar to the - problems encountered with - <code>ACE_Reactive_Strategy</code>, we will likely also - encounter problems with the single - <code>ACE_Threaded_Strategy</code>. Various solutions are - possible, some more generalizable into ACE, other (like the - solution taken with the reactive strategy) localized to TAO. - </dd> - </dl> - - Recently Completed Work: - - <UL> - <li>Broke the tight coupling between <code>CORBA_Object</code> - and <code>IIOP_Object</code> so that multiple - <code>CORBA_Object</code>s can share a single - <code>IIOP_Object</code>. This has a big win in - <code>_narrow</code> operations.</li> - - <li>Eliminated substantial memory leaks in servers.</li> - - <li>Added documentation for <a - href="../Options.html#-ORBpoa"><b>-ORBpoa</b></a>, which allows - individual tweaking of the POA's locality (global - vs. thread-specific) independent of the locality of other ORB - resources. Also, <a - href="../Options.html#-ORBpoalock"><b>-ORBpoalock</b></a> and <a - href="../Options.html#-ORBpoamgrlock"><b>-ORBpoamgrlock</b></a> - control locking in the POA and POA manager, respectively, - according to application needs.</li> - - <li>Began the process of changing the internals of TAO to use - the "underbar" namespace mapping. This should be completed - soon.</li> - - <li>The ORBs manipulation of object keys and object ids is now - consistent. Canonically, everything is managed as an opaque - sequence of octets. Conversion to/from strings is permitted, - with non-printables turning into a trigraph <b>\hl</b> where - <b>h</b> is the high nibble digit in hex, and <b>l</b> is the - low nibble in hex.</li> - - <li><code>CORBA_ORB::run()</code> should be able to be called - multiply in the same execution path now.</li> - - <li>Fixed the <b>LOCATION_FORWARD</b> capabilities to work as - described in the CORBA 2.1 and the POA specification.</li> - - <li>Fixed problem in the IDL Cubit example which caused the - server to dump core when dealing with <code>_out</code> sequence - types <em>ONLY</em> when compiled with G++. It seems that G++ - decided to interpret the expression <code>output == 0</code> - (where <code>output</code> is of type <code>vector_out</code>) - differently from Sun's C++ compiler.</li> - - <li>Fixed a slight problem in <a href="#hostfix">other recent - work</a> where a servant listening on <code>INADDR_ANY</code> - reported an IOR with address <tt>0.0.0.0</tt>. It now reports - the default hostname instead.</li> - - <LI>Working version of TAO on VxWorks with VME transport.</li> - - <li>TAO now reacts gracefully in the face of a non-existent - <code>svc.conf</code> file or simply missing entries in the - file. It logs a warning message and uses default values - versions of items which it would normally obtain from the - Service Repository.</li> - - <li><a name="hostfix"></a>Fixed problem where object references - for objects dynamically created within a server do not have - proper host, at least if the desired host was specified using <a - href="../Options.html#-ORBhost"><b>-ORBhost</b></a> on the - command line.</li> - - <li>Added support for <a - href="../Options.html#-ORBsndsock"><b>-ORBsndsock</b></a> and <a - href="../Options.html#-ORBrcvsock"><b>-ORBrcvsock</b></a> ORB - options. See <a href="../Options.html">options - documentation</a> for details.</li> - - <li>Added support for pre-established connections using the <a - href="../Options.html#-ORBpreconnect"><b>-ORBpreconnect</b></a> - option. This allows connections to be esablished before any - CORBA requests are made. See <a href="../Options.html">options - documentation</a> for details.</li> - - <li>Eliminated deprecated <b>-OAxxx</b> options parsed by the - Object Adapter <code>init</code> method but no longer used by - part of the ORB.</li> - - <li>Made argument parsing within the ORB and POA consume the - arguments it parses.</li> - - <li>Placed an optimization in IIOP::Profile that should speed up - invocations on the client side rather substantially. This was - done by caching an instance of <code>ACE_INET_Addr</code> in - the private portion of <code>Profile</code> because the cost - of constructing one for every invocation (in - <code>TAO_GIOP_Invocation::start</code>) was simply - enormous--the construction was something like 100 cycles on - Solaris. This is all because deep down this ends up calling - <code>gethostbyname()</code>, which can be very costly. Of - course, this is platform-dependent, so the improvements may - not be as great on a different platform.</li> - - <li>Got rid of need to have compile in TSS ORB Core - (<code>-DTAO_HAS_TSS_ORBCORE</code>). This is accomplished by - having a Resource Factory (<code>TAO_Resource_Factory</code>) - singleton which the ORB Core consults in order to get pointers - to resources. The Resource Factory can be configured to - return pointers from thread-specific storage, or - process-global storage. By doing this, the ORB Core is - shielded from ever having to know whether or not its resources - are global or not.</li> - - <li>Made all command-line like options look like the - standard-professed options, i.e., <b>-ORBxxx</b> or - <b>-OAxxxx</b>.</li> - - <li>Moved ORB option parsing into - <code>TAO_ORB_Core::init()</code> rather than in - <code>ORB_init()</code>.</li> - - <li>Define <code>_FAR</code> for all cases as an interim fix - for LynxOS.</li> - - <li>Fixed TAO so that the default port of 5001 is defined in - <code>orb_core.h</code> config header somewhere rather than in - <code>orbobj.cpp</code>, and use <b>MAXHOSTNAMELEN</b> ACE - constant for the hostname buffer.</li> - - <li>Eliminated need for <code>CORBA::Object</code> to keep a - pointer to its orb sitting around anywhere.</li> - - <LI>Tested <code>CORBA_ORB::shutdown()</code> as a manner in - which a thread of control can request that - <code>CORBA_ORB::run()</code> should return and no longer - service requests. It only honors a - <em>wait_for_completion</em> argument of <b>FALSE</b> right - now because we haven't fully explored the issues which - surround doing all the completion.</li> - - <li>Fixed bugs reported by Steven Wohlever, David Miron, - Manojkumar Acharya.</li> - - <li>Created a brief section in <a - href="../../TAO-INSTALL.html"><code>$TAO_ROOT/TAO-INSTALL.html</code></a> - that <em>briefly</em> documents installation on Unix - platforms. There is also a brief section which documents - issues that apply across ALL platforms, such as the fact that - TAO and its application will likely not work well if your ACE - is linked through Orbix, VisiBroker, or another ORB. <a - href="../../ChangeLog">ChangeLog</a> discovery attribution for - this should go to <em>Steve Wohlever - <wohlever@mitre.org></em>.</li> - - <LI>Implemented <code>CORBA_ORB::shutdown()</code> as a manner - in which a thread of control can request that - <code>CORBA_ORB::run()</code> should return and no longer - service requests.</li> - - <LI>Validated movement of Strategy Connector into - ORB Core.</LI> - - <li>Debated details of <code>Svc_Handler</code> modifications - with Irfan.</li> - - <li>Tested and commited <code>Connector::connect_n</code> - changes to ACE</code>.</li> - - <li>ACE & TAO compiling cleanly using Sun's CC on Solaris and - G++ on Linux. Also, the ACE tests run properly.</li> - - <li>Use <code>truss</code> to verify for <a - href="mailto:PAUL.G.WEHLAGE@cdev.com">Paul Wehlage</a> that - TAO really uses <code>select</code> vs. <code>recv</code> for - <b>-R</b> vs. <b>-T</b>.</li> - - <li>Renamed <code>boa.*</code> to <code>poa.*</code>.</li> - - <LI>Renamed following files: Orb_Core.* ->orb_core.* - & orbobj.* -> corba_orb.*.</LI> - - <LI>The lock used in <code>giop.cpp:626</code> originally - appeared in the IIOP-1.4 code base. Its purpose was to - protect access to the <code>fwd_profile</code> data member. - The original features are available, but they are currently - not used because none of our concurrency models present a - thread-unsafe condition. The <a - href="../../ChangeLog">ChangeLog</a> presents more details - on the subject.</li> - - <li>Fixed problems seen by <a - href="mailto:brian.r.mendel@boeing.com">Brian Mendel</a> of - Boeing where the Cubit server application was crashing in - <code>TAO_OA_Connection_Handler::handle_message</code> when it - was trying to call <code>CORBA::POA::handle_request</code> - through a NULL <code>root_poa</code> pointer. The NULL - pointer resulted from the inadvertent use of the - Thread-per-Connection concurrency model, which is currently - incompatible with the ORB-per-thread model. It was not caught - because the change is normally done in configuration files - parsed by the Service Configurator, and VxWorks requires - special code in that area which was not tested until Boeing - got hold of the code.</li> - - <li>Clean up build under g++. All 'function used before - declared inline' warnings have been eliminated. Also, all - other warnings (except for one that is nearly unavoidable) - have been eliminated. - </li> - - <LI>Removed BOA and ROA names from the ORB.</LI> - - <LI>Changed <code>CORBA_ORB::open()</code> to return an error - status (first cut at this already there--needs to be checked - out).</LI> - - <LI>Added README file to multiCubit test.</li> - </UL> - - Ongoing Work: - - <UL> - <LI>Verify ACE & TAO on Linux using g++, Solaris using CC, and - Solaris using g++.</LI> - - <LI>Use Purify to eliminate straggling problems.</LI> - - <LI>Clean up the internals and analyze for performance - bottlenecks using Quantify.</LI> - - <li>Resolve <code>// @@</code> comment blocks. These are - questions or comments placed in the code by a reviewer - (typically Doug).</li> - - <li>Make more and better documentation on the internals of - TAO.</li> - </UL> - - Current Work: - - <UL> - <LI> - <p>Verify that, in the ORB-per-thread concurrency model, - it's possible to have a factory in each thread that creates - a client and returns an object reference to that client - within that thread. This is different from the model that - Sumedh originally had where the factory was trying to turn - out a client in a separate thread/ORB combination.</p> - - <p>There is no application which attempts this at the moment, - so both a client and a server will need to be manufactured. - Using existing Cubit-like examples (that poor app is so - weary) should reduce development time to a couple of - hours.</p> - - <p>The real strategy/solution is to make - <code>CORBA::Object</code> not keep a pointer to its orb - sitting around anywhere. It should only get it when it needs - it. This should not be difficult to change.</p> - </LI> - </UL> - - Critical Work: - - <UL> - <LI>Support new concurrency models, e.g., Thread-per-Request, - thread pool, etc..</LI> - - </UL> - - Future work: - <UL> - <li>Performance optimizations, such as: - <ol> - <li>Strategize the connection management scheme so that we - don't always have to pay the price of a look up in the - connection cache. Basically, we want to be able to "hold" a - service handler for a string of requests. And, of course, - do this in a portable manner.</LI> - - <li>Replace all uses of <code>new</code> and - <code>delete</code> with allocators that can be put into - TSS. The main motivation for this is to eliminate lock - contention when multiple threads need to allocate/deallocate - memory, however it might also be possible to substitute an - allocator better-tuned to an application's specific - usage.</li> - </ol> - </li> - - <li>Robustness improvements: - <ol> - <li>Improve connection cache so that it won't crash when there - are no more slots or sockets.</li> - - <li>Add an option to the ORB core, e.g., - <b>-ORBnameservicetimeout</b>, to allow specification of an - upper bound on the itime-to-wait for a response from the - ORB's multicast "search" for a running name server.</li> - </ol> - </li> - - <li>New features: - <ol> - <li>Implement <code>ORB::perform_work</code> and - <code>ORB::work_pending</code> so that <code>ORB::run</code> - <em>could</em> be implemented in terms of them.</li> - - <li>Improve the <code>Service Configurator</code> service - entry faking on VxWorks.</li> - - <li>Integrate with realtime upcalls (RTUs).</li> - - </ol> - </li> - - <li>Internal restructuring: - <ol> - <li>Remove <code>_FAR</code> from all code.</li> - - <li>Rework underlayment to eliminate <code>IUnknown</code> class.</li> - - <li>Create some sort of internal "request" object which - carries enough context to suport deferred method invocation - such as one might find in the thread-pool approach or - wherever queuing might occur.</li> - - <li>Rethink the entire memory allocation model used by the - ORB. The current one from SSIIOP may no longer be suitable - for our purposes.</li> - </ol> - </li> - - <li>Documentation: - <ol> - <li>Update ORB Patterns paper to reflect the latest TAO has to - offer. Specifically - <ul> - <li>sections on the Abstract Factory and Concurrency - strategies need to be updated, and the entire paper needs to - be reviewed for accuracy w.r.t the current TAO.</li> - <li>Metrics must be re-computed.</li> - <li>Tighten existing discussions.</li> - <li>Add information on new patterns (such as TSS) which - have been used since initial work.</li> - </ul> - - <li>Make the section in <a - href="../../TAO-Install.html"><code>$TAO_ROOT/TAO-Install.html</code></a> - that documents installation on Unix platforms more - complete.</li> - </li> - - </ol> - </li> - - <li>Miscellany: - <ol> - <LI>Assist in porting higher level services (e.g., <A - HREF="#eventchannel">Event Channel</A>) to TAO.</LI> - - <li>Compile and verify ACE and TAO on Linux using the Kuck and - Associates, Inc. (KAI) compiler tools.</li> - <LI>CORBA::TypeCode is causing a memory leak. Since objects - of this type are rather static in nature, this is - currently not a big deal. However, it should be fixed - soon. </LI> - <LI>CORBA_Exception::Release is not working correctly as the - memory is never really freed. </LI> - </ol> - </li> - - - <li>Potential Optimizations: - <ol> - <LI>There is a certain amount of overhead introduced by the - use of TSS Singletons. TSS singletons are very - convenient as the variable does not have to be passed - from function to function. Even though sort of access is - generally acceptable, the overhead of this approach - maybe too high when used in code which is executed - frequently (e.g., during an upcall). This critical code - may have to be restructured not to use TSS Singletons - because of the high overhead associated with it. </LI> - </ol> - </li> - - - </UL> - - </body> -</html> |