IDL Compiler

Current status: (As of April 30, 2001.)

• Generated code closely follows the C++ Mapping specified in the latest C++ mapping for CORBA 2.3 (Document ptc/98-09-03).
• IDL compiler is now able to generate code that support native C++ exceptions on the stubs and skeletons. With this strict mapping, the CORBA::Environment parameter is no longer generated. The default behavior is to generate code without the extra parameter on plaforms with native exceptions and with the extra parameter in platforms without native exceptions. Use the -Ge flag to override the defaults.
• We are now able to handle shared case labels and default label in unions. In addition, whenever appropriate, we are also able to generate the "default ()" operation.
• We are now able to handle recursive types. We are also able to generate optimized typecodes.
• Struct members of type strings and arrays of strings now use the managed type instead of the _var type. This change was necessary to conform to the IDL->C++ mapping.
• Fixed a large number of problems with anonymous arrays and sequences inside structs and unions. The name of anonymous sequence needs to be fixed as per latest C++ mapping spec.
• Compile problems with sequence of forward declared interfaces is fixed. In addition, problems with sequence of CORBA::Objects is fixed. In this specific case, we were not generating the _downcast and _upcast methods.
• Some more problems with the front-end have been fixed. In particular, oneway operations with a "raises" clause or having an "inout", "out", or "return" mode is flagged as an error.
• For platforms that support namespaces, we now allow reopening modules.
• Support for generating compiled marshaling code is added. Use the -Gc option. However, this needs thorough testing before we can claim success. Unions are still a problme with compiled marshaling.
• The problem of "#include"ing the relative path of the header files rather than the paths of their corresponding IDL files has been fixed. tao_idl now generates #include path names that are derived from the IDL files that are #include'd in the main idl file.
• Added options to IDL compiler to specify file name endings for the IDL-generated stubs, skeletons and the various header files. Please refer to the IDL compiler options for details.
• Verified support for the "long long" and "unsigned long long" datatypes. On platforms that do not support 64 bit longs we provided partial emulation through ACE_U_LongLong.
• Perfect Hashed Operation Lookup Strategy has been added to the IDL Compiler. -P flag to tao_idl enables the perfect hased lookup strategy. This strategy uses GPERF, the GNU's Perfect Hash Function Generator written by Dr.Douglas C. Schmidt. Right now, GPERF works only on Solaris. Any work on porting GPERF to other platforms will be highly appreciated.
• The <<= and >>= operators for user-defined types are now generated.
• Completely redesigned the IDL compiler using the Visitor patterns. Many incomplete issues have been resolved. These include support for "sequence of typecodes", passing object references as in, inout, and out parameters. Code generation for sequences is also properly handled i.e., for a named sequence such as typedef sequence<char>CharSeq;, we now generate a new class (and hence a type) called "class CharSeq". Arrays are still being worked out and will be done soon. An important difference in the generated code is that the skeletons now use a table driven approach very similar to the stubs.
• Support for the "native" keyword added.
• The problem of incorrect code generation for typedefs defined in an imported file is resolved.
• Problems when interfaces use single or multiple inheritance solved. The problem was with the demultiplexing code, the generated operation tables, and the dispatching mechanism. We are currently testing this with the Event Channel code.
• The problems arising due to public virtual inheritance when casting from an interface class to CORBA::Object_ptr has been solved. We do this casting inside the stubs/skeletons rather than first converting an interface class pointer to a void*, storing it in an Any, and casting it to CORBA::Object_ptr in the encode/decode methods. The casting inside the stubs/skeletons work because the compiler has knowledge of both types.
• Include files are handled properly. So are the definitions used inside the include files that are used in the currently parsed files.
• Generates C++ stubs and skeletons that use TAO's interpretive IIOP protocol engine.
• Support dynamic libraries on NT, i.e., marking classes for DLL export was added. Two backend options control the name of the export macro, and the name of an extra include file were the macro is defined; the options are -Wp,export_macro=MACRO_NAME-Wp,export_include=INCLUDE_NAME.
• The IDL compiler generates now source code for sequences. The user has now the option to use these generated sequence classes or to use, as up to now, the template instatiation. If TAO_LACKS_TEMPLATE_SPECIALIZATION is defined, then template instantiation will be used, else not. The reason for this was, that some C++ compilers did not support template instantiation properly and sequences were based on templates. The generated source code is mainly contained in the generated header file directly in the class declaration.
• The IDL Compiler generates templates for servant implementations. The options are -GI [ h | s | b | e | c ]
• The IDL compiler generates source code for the management and (de)marshaling of wide characters and wide strings, enabling the sending and receiving of Unicode over the wire. However, wide character and wide string literals are not yet portable to Unix platforms (see entry under Future Work below).
• Since the CORBA spec requires that all enums be 32 bits, and some compilers will try to use less space if the enum values are small enough, the IDL compiler now appends _TAO_ENUM_32BIT_ENFORCER = 0xFFFFFFFF to every enum. This appended enum value is not part of the IDL compiler's internal representation of the enum, so unions that use the enum as a discriminator will not have incorrect _default() code generated for them.
• The IDL compiler generates a C++ ostream operator for IDL exceptions. So far only the repository ID is output, but this may be enhanced when requirements and/or desires become clearer.
• The IDL compiler now has limited support for valuetypes (see release notes for valuetypes for details). If the TAO library is built with TAO_HAS_VALUETYPE defined, and IDL_HAS_VALUETYPE is also defined, then the IDL compiler will enable OBV support with the command line option -Gv, and disable it with the option -Sv (the default).
• As part of the implementation of interceptors, the TAO IDL compiler now generates interception points in the client and server, as well as the prepare_header method in the stubs.
• Scoping and name resolution rules have changed in CORBA with version 2.3. The IDL compiler now conforms to these new rules.
• IDL compiler now supports the CORBA AMI callback model, generating code for reply handlers and reply stubs if the -GC command line option is used. The TAO library must be compiled with TAO_HAS_CORBA_MESSAGING = 1. If this is done, TAO_HAS_AMI_CALLBACK will automatically be defined to 1 as well. IDL_HAS_VALUETYPE is defined to 1 by default.
• New command line option -So added to suppress generation of ostream operators for exceptions.
• New command line option -Sc added to suppress generation of tie classes and *S_T.* files. The default is still to generate them.
• IDL compiler now handles escaped identifiers (CORBA 2.3.1). An identifier appearing in an IDL file with a leading underscore will appear in generated code without the underscore. This enables the use of identifiers in generated code identical to IDL keywords, as specified in CORBA 2.3.1. If the resulting identifier matches a C++ keyword, "_cxx_" will be prepended in generated code as before.
• The -St option to suppress generation of typecodes now also suppresses the generation of the Any insertion and extraction operators. The extraction operators require the associated typecode, so the generated code for those operators would not compile when the -St option was used.
• Option -Ge 2 added which generates 'throw' instead of ACE_THROW_SPEC, ACE_THROW, and ACE_RETHROW. Since the expansion of ACE_THROW_RETURN is platform-dependent, it was left as is. Same for TAO_INTERCEPTOR_THROW, since it sometimes expands to ACE_THROW_RETURN. Of course ACE_HAS_EXCEPTIONS must be defined for this option to work.
• Removed generation of ostream operators for user exceptions, along with the command line option to suppress this generation. The ostream operator defined in the base class CORBA::Exception works fine and produces the same output.
• The TAO IDL compiler is no longer monolithic. It is composed of a top-level executable, a front end library and a back end library. This will enable different back ends to be plugged in for code generation in different languages, and for IfR administration. Different back ends may require a few changes to the executable (described below), but the front end library can remain unchanged. The chain of dependencies is as follows:

  FE : ACE
  BE : FE
  EXE : FE, BE
  

  Executing the Makefile (or the TAO_IDL Compiler project in the MSVC 'tao_idl' workspace) will build whatever is required in the proper order, as before. Back end files, classes and functions required by the executable are as follows:

  be.h - file containing #includes of the major BE file headers.
  TAO_Codegen - class, holds output stream references for the various generated files.
  TAO_CODEGEN - ACE_Singleton typedef of the above class.
  tao_cg - pointer to instance of TAO_CODEGEN.
  be_generator - class, inherits from AST_Generator in FE and generates AST nodes.
  void BE_produce (void) - global function, starts AST traversal for code generation.
  BE_GlobalData - class, holds default/command line arg settings specific to BE.
  be_global - pointer to instance of above class.
  

  All the code in the executable that may need to be modified for different back ends in contained in the file drv_args.cpp. Code in this file processes the command line arguments, outputs a usage message, and performs other miscellaneous BE initialization.
• The IDL compiler can how handle interfaces forward declared in one IDL file and defined in another. It will handle interfaces that are mutually dependent across two IDL files, as long as code generated from both IDL files is included in the same C++ build.
• Generation of template tie class declarations has been moved from the *S.h file to the *S_T.h file. The new SunCC 5.2 compiler does not require that template source code be included in the header file, but it does require that template declarations and implementations be in the same 'compilation unit' that is, the same header/source file combination. Generated ifdef guards prevent the compilation of tie class code if the target platform does not support namespaces. For such platforms, TAO maps the IDL 'module' keyword to a class instead of to 'namespace', and C++ does not allow template class declarations to occur inside another class.
• IDL compiler can now process multiple IDL files per execution, on all platforms. A separate process is spawned for each file. IDL files and command line options may appear in any order on the command line. Any option not starting with '-' will be treated as a filename.

Known bugs/unimplemented constructs:

Goal is to measure the code size of the interpretive stubs generated by TAO IDL compiler vs code size of compiled stubs. Then compare the performance of each. We want to prove the thesis that TAO IDL compiler generated interpretive stubs have a small code size, yet are comparable in performance (or slightly less) than compiled stubs. Hence, it will be useful for small distributed equipment such as handsets, PDAs, etc.

In doing the above, improvements to the IIOP protocol engine in terms of size/performance/determinism will be made.

• Tweak the IDL compiler to generate code that's more easily integrated back into the ORB Core, e.g., POA, etc. This will depend largely on our ability to generalize the changes necessary to generated code.
• The generated sequence classes should not be generated per sequence, but per type and parent scope. Which means, that the overhead of having the source code generated serveral times should be reduced. To do this, an extra pass over the internal representation of the IDL file has to be done.
• Unix's native wchar_t type is not compatible with Unicode. CORBA wide string implementations are required to support Unicode, but are not restricted to it, so the IDL compiler should ultimately support Unix's wchar_t as well.

Pluggable Protocols

The goal of the pluggable protocol effort is to (1) identify logical communication layers in the ORB, (2) abstract out common features, (3) define general interfaces, and (4) provide necessary mechanisms for implementing different concrete ORB and transport protocols. TAO's pluggable protocol framework will allow disparate communication mechanisms to be supported transparently, each with its own set of requirements and strategies.

For example, if the ORB is communicating over a system bus, such as PCI or VME, and not all the features of GIOP/IIOP are necessary and a simpler, optimized ORB and transport protocol can be defined and implemented. Similarly, it should be straightforward to add support for new transport protocols that use native ATM or shared memory as the underlying communication mechanism. In all cases the ORB's interface to the application will remain compliant with the OMG CORBA standard.

There will be several stages of the development process: (1) basic pluggable transport protocols framework, (2) support for multiple profiles, (4) add example transport protocols, such as ATM and VME, and refine/optimize the transport protocols framework, and (4) add support for pluggable ORB protocols, e.g., replacements for GIOP. Each of these steps is outlined below:

• Basic pluggable transport protocols framework: We have added several Bridge classes that decouple the transport-specific details from the rest of TAO's ORB Core. This allows us to isolate the details of how messages are communicated at the transport layer in a few classes. This design resulted in the restructuring of the ORB Core and how requests are handled. For instance, there is now the concept of communication layers: Objects (e.g., references, method invocations, etc.), ORB Messaging, Transport, and Network. The Object layer is just the usual stubs and skeletons.

The common interfaces have been defined in the new abstract classes that form the core of TAO's pluggable protocol framework, e.g., TAO_Connector, TAO_Acceptor, TAO_Profile and TAO_Transport. Two new mechanisms for keeping track of supported transport protocols are the TAO_Connector_Registry and TAO_Acceptor_Registry, which are essentially Abstract Factories that produce the right types of connector, acceptors, and transports.

• Multiple Profile - Support for more than one profile per object. This is important since there may be several different ways to access an object. Each profile for an object may encode information pertaining to QoS, network and transport protocols, addresses or routes.

• Example Transport protocols - Aside from IIOP, four additional transport protocols are distributed with TAO:

  1. UIOP: GIOP over local IPC (UNIX domain sockets)
  2. SHMIOP: GIOP over shared memory
  3. SSLIOP: IIOP over SSL (Secure Socket Layer)
  4. DIOP: GIOP over UDP/IP

  Other interesting transport protocols could be for ATM, Buses (VME or PCI), TP4, and GSMP. TAO users have also created their own pluggable transport protocols, such as a ScramNet pluggable protocol.

• Pluggable ORB protocols - This step will add support for ORB protocols besides GIOP. In particular, we will explore lightweight protocols using shared memory and system buses like PCI or VME.

• The basic framework to support pluggable transport protocols has been completed. The standard TAO regression tests Latency, MT_Cubit, Multiple_Inheritance, CDR and EC_Throughput can be used to verify performance using the new framework.
• Multiple endpoint support in the ORB has been added. A list of TAO_Acceptors is kept in the Acceptor Registry. When the ORB needs to create an IOR it iterates over all the acceptors to do so. Using either multiple -ORBEndpoint options or several endpoints separated by semi-colons ';', the user can specify what addresses the ORB should use. Each endpoint is specified in URL format (ex: iiop://foo.bar.com:0), this format can be extended to support different protocols.
• If the user does not specify a list of endpoints then the ORB creates a default endpoint for each protocol configured, unless the pluggable protocol explicitly prevents that in an effort to prevent resource leaks.
• Added support for multiple Connectors in the ORB, the ORB finds the correct connector based on the tag for the profile.
• Added support for multiple profiles in the IORs, when the ORB demarshals an IOR it queries the Connector Registry to create the right kind of profile for the known protocols. If one of the protocols is unknown we create a special profile class that can only be used for marshaling and demarshaling, not communication.
• Enabled the UIOP protocol. This protocol uses local IPC (aka UNIX domain sockets) as the transport mechanism. The protocol is loaded by default.
• Enabled the SHMIOP protocol. This protocol uses shared memory as the transport mechanism. The protocol is loaded by default.
• An IIOP over SSL pluggable protocol called "SSLIOP" has been implemented. It provides secure communication between hosts that support IIOP over SSL, and is meant to be a drop-in replacement for the IIOP pluggable protocol. TAO's SSLIOP pluggable protocol implementation supports both the standard IIOP transport protocol and the secure IIOP over SSL transport protocol.

  No changes were made to the core TAO sources to provide to this SSL support, nor does TAO contain any security related hooks. TAO's SSLIOP implementation is completely self-contained. This ensures that the core TAO sources remains free of export restrictions.

• Protocols can be dynamically loaded into the ORB: The default resource factory reads the protocol "names" from its list of arguments. These protocol names are used to load an abstract factory via the service configurator. This factory can create acceptors or connectors on demand. By default only IIOP is loaded.
• The service configurator is now used to load protocol factories.
• Support for the -ORBHost and -ORBPort has been removed. The new -ORBEndpoint option supersedes them, and provides the same functionality in a protocol-neutral way. If the deprecated options are used, the ORB exits with a CORBA::BAD_PARAM exception, indicating an unknown -ORB option.
• The -ORBPreconnect ORB option has been deprecated in favor of the standard validate_connection run-time feature. Support for this option will be removed from future releases.
• The URL style object reference format has been updated to conform with the format that corbaloc uses. The BNF specification for corbaloc is:

  <corbaloc> = "corbaloc:/"[<obj_addr_list>]["/"<key_string>]
<obj_addr_list>= [<obj_addr> ","]* <obj_addr>
<obj_addr>= <prot_addr> | <future_prot_addr>
<prot_addr>= <rir_prot_addr> | <iiop_prot_addr>
<rir_prot_addr>= <rir_prot_token>":"
<iiop_prot_addr>= <iiop_id><iiop_addr>
<iiop_id>= ":" | <iiop_prot_token>":"
<iiop_prot_token> = "iiop"
<iiop_add> = [<version> <host> [":" <port>]]
<host> = DNS-style Host Name | ip_address
<version> = <major> "." <minor> "@" | empty_string
<port> = number
<major> = number
<minor> = number
<future_prot_addr> = <future_prot_id><future_prot_addr> <future_prot_id> = <future_prot_token>":" <future_prot_token> = possible examples: "atm" | "dce" <future_prot_addr> = protocol specific address <key_string> = <string> | empty_string


  The uiop URL style object references syntax is:

  <uioploc> = "uioploc://"[<addr_list>]["|"<key_string>]
<addr_list>= [<address> ","]* <address>
<address> = [<version> <rendezvous point>]
<rendezvous point> = Valid Filesystem Path
<version> = <major> "." <minor> "@" | empty_string
<major> = number
<minor> = number
<key_string> = <string> | empty_string


  Note that the key string delimiter for uiop is a vertical bar `|' (the command line "pipe" symbol) not a forward slash `/'. A delimiter other than a forward slash is needed to prevent ambiguities of where the rendezvous point ends and where the key string begins since both may contain forward slashes in them. The new corbaloc:uiop URL format may also be used.

  It should be noted that these formats have been superseded by the new corbaloc Interoperable Naming Service support.

• The rendezvous point for uiop is any valid path and filename that the ORB has permission to read and write to. However, UIOP rendezvous points have the same restrictions that local IPC has. The following are some guidelines that will help ensure successful use TAO's UIOP pluggable transport protocol:

  To guarantee portability, local IPC rendezvous points (including the path and filename) should not be longer than 99 characters long. Some platforms may support longer rendezvous points, usually 108 characters including the null terminator, but Posix.1g only requires that local IPC rendezvous point arrays contain a maximum of at least 100 characters, including the null terminator.

  If an endpoint is longer than what the platform supports then it will be truncated so that it fits, and a warning will be issued.

  Avoid using relative paths in your UIOP endpoints. If possible, use absolute paths instead. Imagine that the server is given an endpoint to create using -ORBEndpoint uiop://foobar. A local IPC rendezvous point called foobar will be created in the current working directory. If the client is not started in the directory where the foobar rendezvous point exists then the client will not be able to communicate with the server since its point of communication, the rendezvous point, was not found. On the other hand, if an absolute path was used, the client would know exactly where to find the rendezvous point.

  It is up to the user to make sure that a given UIOP endpoint is accessible by both the server and the client.

  It is important to be consistent in the use of absolute paths and relative paths for rendezvous points. The two types of paths should not be used for the same endpoint. For example, if uiop:///tmp/foo is specified as the server endpoint and uiop://foo as a preconnect for a client in /tmp, then the preconnection may be established but it is likely it won't be used since the endpoint and preconnect are interpreted as different strings, i.e. /tmp/foo and foo are not the same, lexicographically. On the other hand, if both the endpoint and the preconnect are the same string then a preconnection will be established and used successfully.

  The -ORBEndpoint option uses a syntax similar to that of the URL style object reference shown above. The only difference is that the object key delimiter and the object key string are not specified.
• Added documentation that describes how to implement pluggable transport protocols for TAO. The document is available here.
• TAO's IIOP pluggable protocol now supports automatic creation of profiles for endpoints created on a host with multiple network interfaces. It should no longer be necessary to manually specify an endpoint for each network interface.

  This means that server IORs will contain profiles for all of the default endpoints created on each network interface the server is listening on, if no explicit endpoints were specified.

Known Issues:

• None.

• Complete support for multiple ORB messaging protocols.
• Long term work will include adding support for pluggable ORB protocols, as well as transport protocols. This way we can develop optimal messaging and transport protocols for a given platform.

Portable Object Adapter (POA)

The POA associates servants with the ORB and demultiplexes incoming requests to servants.

Current Status:

• TAO supports the POA spec. This section will carry updates as available.

• ORB::shutdown now properly deactives all the POA Managers.
• POA Managers in TAO were previously ignored in the request processing path on the server. This is now fixed such that their state is checked before dispatching the client request to the servant. Only if the state is ACTIVE, is the request dispatched to the servant. Otherwise, the request is rejected. Since POA Managers start off in HOLDING state, make sure to activate() them before falling into the event loop.
• TAO's POA now properly supports both the threading policies: SINGLE_THREAD_MODEL and ORB_CTRL_MODEL.
• The synchronization in the POA is now very optimal. For example, the locks are not held across the invocation on the servant. The locks are also not held across the invocation on the AdapterActivator and ServantManagers. This allows us to use regular locks instead of recursive locks inside the POA. This also allows multiple threads to dispatch requests on the same POA simultaneous.
• TAO now supports reference counting between POA and servants, including the new RefCountServantBase and ServantBase_var classes. RefCountServantBase is a reference counted base class that was added to the CORBA specification to avoid race conditions for servant deletion in threaded servers. ftp://ftp.omg.org/pub/docs/orbos/98-07-12.pdf contains the relevant text. Check here on some hints to avoid trouble.
• The POA now supports active demultiplexing of servants in the SYSTEM_ID and the USER_ID policy. This should make the POA faster and more predictable since there is no hashing involved and the index of the slot where the servant is registered is in the Object Key.
• Previously, the complete POA name was used as the POA identity. This scheme was inefficient in many ways including: (a) the complete POA name can be significantly large in size, and therefore, ineffient to pass with every method call from the client to the server; (b) it is varible in size, and therefore, does not lend itself to smart and effective parsing; (c) the searching based on the complete POA name is very ineffient.

The new solution is to use an active demux table, and flatten the POA hierarchy. This will help in the searching since active demuxing is fast and predictable. This will also help in the parsing since the demux key will be fixed size.

Note that for persistent ids, we have to pass the complete POA name in addition to the demux key in order to handle POA creation on demand.

• There were some POA objects in a typical server that are not freed up properly, resulting in a memory leak. This has now been fixed.
• Timestamps in persistent IORs were not required and have been removed.
• POA exceptions are not not system exceptions and have been removed from the list of system exceptions.
• Vastly improved the ability of the POA to deal with user exceptions, memory allocation failures, and constructor failures.
• We now support a minimum POA for the minimum CORBA specification. Recently, this feature was enchanced such that the minimum CORBA footprint was further reduced. In addition, minimum POA can be enabled/disable irrespective of the minimum CORBA setting.
• We have decided not to support active demuxing for method name lookup. The benefit of this optimization was questionable since the current perfect hashing scheme provide very good and predictable behavior.

Also, note that this optimization will require many changes. We would have to use the help of the IDL compiler to modify the object key that is passed for every method call differently. Note that this scheme doesn't work in the case of multiple inheritance or when the client stubs are not TAO.

• Improved the parsing of object keys belonging to the RootPOA. Since this is the default POA and is commonly used, we have given it a reserved byte in the object key in order to quickly identify it. With the reserved bit, the active demux key for the RootPOA is not used, and no map lookups are required.
• POA name separator was changed from '/' to '\0'. Since POA names are strings, this makes a better choice since there is no chance of a conflict with the string specified by the user.
• We have support for reactivating servants with system generated ids.
• The TAO specific synchronization POA policy has been removed.
• New examples have been added to show how servants can be dynamically loaded from DLLs on demand.
• Support for collocation should be much better now because the POA can tell if we created the object reference.
• After Nanbor's recent changes for collocation, we support the full semantics of remote objects on a collocated object. The spec mandates that collocated object should behave exactly like remote objects, which includes going through the POA, running the Servant Managers, running the interceptors, and expecting the reference counting behavior provided by the POA. Note that the old scheme of direct call through to the servant is also still available.

CORBA Naming Service and Interoperable Naming Service

OMG defined CORBA Naming Service (spec here) to provide a basic service location mechanism for CORBA systems. CosNaming manages a hierarchy of name-to-object-reference mappings. Anything, but typically the server process hosting an object, may bind an object reference with a name in the Naming Service by providing the name and object reference. Interested parties (typically clients) can then use the Naming Service to resolve a name to an object reference.

More recently, CORBA Naming Service was subsumed/extended by the CORBA Interoperable Naming Service, a.k.a. INS (spec here). INS inherits all the functionality from the original Naming Service specification in addition to addressing some its shortcomings. In particular, INS defines a standard way for clients and servers to locate the Naming Service itself. It also allows the ORB to be administratively configured for bootstrapping to services not set up with the orb at install time. This page provides a brief description of additional features INS provides, and how they are implemented in TAO.

Current status (as of 18 May 2001):

• The Naming Service now does not do listen for multicast discovery as the default. Use -m1 option to turn it on.

• Changed default for multicast discovery to

• Added support for Persistence (using memory-mapped files). Persistence feature is optional, and is controlled by the command line argument.
• Updated the implementation of the Naming Service to use new ACE exception macros.
• Added support for the InterOperable Naming Service, which enables the ORB to support IORs in user-friendly corbaloc format. These features allow the ORB to be configured to return arbitrary object references from CORBA::ORB::resolve_initial_references for non-locality-constrained objects. Two options -ORBInitRef and -ORBDefaultInitRef have been added to the orb for bootstrapping to arbitrary services.
• Added support for the Naming service to act like an agent: to understand IIOP request messages from clients and respond with reply messages with a LOCATION_FORWARD/OBJECT_NOT_EXIST status. The Naming Service can be configured through ORB options to register arbitrary services given the URL-format IOR for the service. The resolve_initial_references () resolves a service in the following order :
  1. -ORBInitRef
  2. -ORBDefaultInitRef
  3. Multicast discovery
• Added a test for the InterOperable Naming Service that works in conjunction with the current TAO examples.
• Implementation of the CORBA INS spec is completed. The following features are now supported:
  • corbaname format
  • NamingServiceExt interface

• Support for a load balancing feature similar to the one present in ORBIX. It will be possible to bind a group of objects under a single name, and when a client attempts to resolve the name in question, a preset policy (e.g., random, round robin, etc.) will determine which one of the object references from the group will be returned.
• Support for the Naming Service to handle the IIOP LocateRequest messages and respond with LocateReply messages with a LOCATION_FORWARD/OBJECT_NOT_EXIST status.

CORBA Trading Service

The Trading Service is an implementation of the COS Trading Service speficiation that meets the Linked Trader conformance criteria --- it implements the Lookup, Register, Admin, and Link interfaces, but not the Proxy interface. Notably, the TAO trader supports the following features:

• Multithreaded operation;
• Trader federations and distributed queries;
• Dynamic properties;
• Modifiable properties;
• All policies described in the specification;
• Preference sorting;
• Service type inheritance hierarchies and subtype searching.

Future Work:

• The Proxy Interface.
• Persistent storage of service types and offers.

CORBA Property Service

Current status (as of Mar 9th, 1999): All the interfaces of this service have been implemented. Please go through the test examples at $TAO/orbsvcs/tests/CosPropertyService. Property Service is has been used by the TAO's Audio Video Streaming Servicedeveloped for TAO. For general documentation of the Property Service, please read The Property Service Specification.

Recent Work:

• Changed the PropertyException from Exception to struct, according to the OMG's changes.
• Changed the implementation to allocate storage for the Sequence out parameters, eventhough their length is 0. This is according to the CORBA specification.

CORBA Concurrency Service

Current status (as of May 3rd): The Concurrency Service provides a mechanism that allows clients to acquire and release various types of locks in a distributed system.

• A simple version of the Concurrency Service has been implemented, i.e. a version without transactions. It is currently being tested.

• Implementation of the Concurrency Service with transactions

CORBA Audio/Video Streaming Service

This is an implementation of the OMG spec addressing the Control and Management of Audio/Video Streams.For more documentation on TAO's A/V Service please have a look here.

Current Status:

• The audio/video streaming service has been implemented in the full profile. The current implementation support all the flow related components like flowEndpoint,FDev,FlowConnection,..,etc.
• Point-to_Point and Point-to-MultiPoint streams have been implemented.
• A Pluggable protocols framework has been implemented to flexibly add new flow protocols like SFP, RTP and new transports like ATM. The current implementation has protoocol implementations for SFP, RTP over UDP and Multicast UDP. Please look at this paper for more documentation about the implementation.
• A Videoconferencing application based on Vic, a MBONE tool, has been implemented using the AV components. Please contact yamuna or Naga if you're interested.
• An MPEG-1 application which streams mpeg-1 video and mpeg-1 audio separately has been developed using the service. The client side of the mpeg player requires X windows support.Its available in the release at $TAO_ROOT/orbsvcs/tests/AVStreams/mpeg/source.
• An Integrated Video-on-demand application has been developed using the Trading Service and the A/V Service. The demo uses a Java FrontEnd and JNI to talk to the TAO C++ trader client. The demo is available in the release at $TAO_ROOT/orbsvcs/tests/AVStreams/server_discovery.

Work in progress:

• Adding more features to the VideoConferencing application eg. porting vat, the audio counterpart of vic to use AV components.
• Adding ATM AAL5 protocol support.
• Adding QoS implementation.

CORBA Time Service

Point of contact: Vishal Kachroo

The Time Service allows clients to connect to Time Service Clerks and obtain globally synchronized time. This time is calculated from the time obtained from one or more Time Servers running on multiple machines in the network. The service uses the TAO Implementation Repository to activate the time servers on demand.

Current status (as of 10th Jan 1999):

• Implementation of a Distributed CORBA Time Service is complete.

• Currently the average of the time obtained from the various servers is considered the global notion of time. A better distributed time synchronization algorithm can be used in the future.
• Implementation of the Timer Event Service.

CORBA Event Service

Last updated: Fri Mar 5 20:38:26 CST 1999

The COS compliant Event Service implements the Event Service Specification: (.pdf), (.ps)
This implementation is based on the Real Time Event service.

Features in this release:

• There is a new implementation of the COS Event Service available. This new implementation supports both the Push and Pull styles for event communication, and it does not require the Real-time Event Service to work.

• A new testsuite for the COS Event Service has been started, they are available at: $TAO_ROOT/orbsvcs/tests/CosEvent/

• A new example for the COS Event Service is provided in $TAO_ROOT/orbsvcs/examples/CosEC/Simple/

• A new binary to run the native COS Event Service was added, it is compiled in: $TAO_ROOT/orbsvcs/CosEvent_Service/CosEvent_Service_Native

• A simple test ($TAO_ROOT/orbsvcs/tests/CosEC_Basic) demonstrates how to create and use the event channel.
• Event Service ($TAO_ROOT/orbsvcs/CosEvent_Service)The Event Service creates a COS compliant event channel and registers it with the naming service with the default name "CosEventChannel".

Please read the associated README for more details.
• CosEC_Multiple: ($TAO_ROOT/orbsvcs/tests/CosEC_Multiple): This test demonstrates how multiple CosEC's connect to one RtEC and how multiple consumers and producers exchange events in this configuration.

Known bugs:

• CosEC_Multiple: ($TAO_ROOT/orbsvcs/tests/CosEC_Multiple): Once the tests are done, the control doesn't return to the shell, you have to say CTRL-C to get back to the prompt.

CORBA Telecom Log Service

Last updated: Sat Feb 19 16:58:42 CDT 2000

The CORBA Telecom Log Service prototype was released in TAO version 1.0.6.

Features supported in the first version:

• The Log Service implementation under $TAO_ROOT/orbsvcs/orbsvcs/Log implements the DsLogAdmin module.
  
• The Logging_Service ($TAO_ROOT/orbsvcs/Logging_Service) starts the Log Service and registers the Log Service Factory with the Naming Service as "BasicLogFactory".
  
• The test $TAO_ROOT/orbsvcs/examples/Log/Client demonstrates a simple use of the Log Service.
  A thorough test $TAO_ROOT/orbsvcs/examples/Log/test/client tests most of the features of the logging service.
• The Query Language supported at present is "TCL". The Log record id and Log record time are treated as properties of the log and can be referenced in a query as "id" and "time" respectively.
  

Things to be done:

• Implement a few remaining methods -
  For the Log interface: copy.
  For the LogMgr interface: list_logs_by_id
  

Future work and enhancements:

• Support the DsEventLogAdmin module, which uses the COS Event Service
  
• Support the DsLogNotification module after the CORBA Notification Service is complete.
  
• Currently all Log records are stored in memory, later we could add support for persistant storage.
  
• Support "Extended TCL" as the Query Language.
  Support the record attributes as name-value pair properties.
  
• Use Red-Black trees to optimize lookup on frequently used query keys - namely record id's and time.
  

CORBA Notification Service

Last updated:Wed Feb 14 15:14:15 CST 2001

Work is in progress to implement the CORBA Notification Service . The implementation released in TAO 1.1.12 consists of the following (see the associated README's for more information):

• The implementation of the interfaces in the Notification Service spec is in $TAO_ROOT/orbsvcs/orbsvcs/Notify.
• The service executable is implemented in $TAO_ROOT/orbsvcs/Notify_Service.
• The example in $TAO_ROOT/orbsvcs/examples/Notify/Filter shows a basic example of using filters.
• The example in $TAO_ROOT/orbsvcs/examples/Notify/Subscribe shows a basic example of how to use subscriptions.
• There is on-going work to add tests in $TAO_ROOT/orbsvcs/tests/Notify/

Features supported thus far:

• The following QoS properties are supported - per-message event priority,order policy and discard policy.
• The following Admin. properties are supported - max. queue length, max_consumers, max_suppliers and reject_new_events.
• All Proxy objects support the Subscribe and Publish interface methods and will inform connected clients about subscription_change and offer_change.
• Assignment of ID's: All object factories assign ID's to objects created.
• For now, the SequenceProxyPushSupplier does not push to consumers. This will be done once we support the Maximum Batch Size QoS property.
• The EventChannel's default Filter Factory can be used to create filters. For now we support the "TCL" grammer from the Trading Service.
• This implementation is backward compatible with the CORBA Event Service.
• A runnning theme in this project is to isolate the patterns identified in the implementation of the RTEC and reuse them. We hope to come up with a collection of C++ template classes that capture these patterns so that they might be used in other similar event communication senarios.
• Concurrency:
  • The service is thread safe.
  • MT dispatching of events to consumers - The EventChannel will deliver the event in another thread, different from the thread that evaluated (checking subscriptions and filters) it.
  • MT Filter evaluation of events - Events delivered to the ProxySuppliers will be evaluated in another thread, different from the delivering thread.
  • MT dispatching of updates - subscription/offer_change updates are delivered on another thread.
• Memory Management
  • (almost) All memory allocations are via factory interfaces.
  • Uses service configurator to load resource factories and supply strartup options to the service.

Project Schedule

• Implement the QoS properties - Expiry times, Earliest delivery times, maximun events per consumer, deadline based ordering, maximum batch size, pacing interval.
• Filters:
  • Intergrate the parser engine for the Extended TCL grammer in orbsvcs/orbsvcs/ETCL with Notify and remove dependency on the Trader .
  • Implement CosNotifyFilter::MappingFilter.
  • (Wish list) Add a "filter cache" to cache the expression tree for frequently used constraint expressions.
• Come up with a rigorous test suite for this service to measure its performance.
• Implement Pull interfaces.
• Support policies to remove dead or unresponsive consumers and suppliers in the Notification Service.

CORBA Security Service

For up-to-date information, see the project web site here.

Implemented Features:

• IIOP over SSL integration via TAO's SSLIOP pluggable protocol.

  • Added an SSLIOP::Current implementation that can be used to obtain the SSL session state for the current execution context. This is useful for obtaining the SSL peer certificate chain associated with the current request, for example.
  • TAO's SSLIOP pluggable protocol now registers a secure invocation server request interceptor. It enforces secure invocation by rejecting requests coming in on the insecure port if the server is configured to do so (default behavior).
  • Implemented SecurityLevel1 for the SSLIOP security mechanism.
  • The Security::QOPPolicy policy has been implemented. This policy is used to set the desired invocation Quality-of-Protection (QoP). It can be created using ORB::create_policy(), and used in conjunction with the standard policy manipulation CORBA features (e.g. PolicyManager, PolicyCurrent), meaning that this policy can be set on a per-ORB, per-thread or per-object basis.

    This policy makes it possible to, for example, make both secure and insecure invocations within the same client process.

• Core SecurityLevel2 interfaces such as Credentials are currently under development.

• Septempter 2001
  The following milestones establish a strong foundation upon which the remaining, more "interesting," CORBA Security Service features will be built on. Once they are in place, it will also be possible to perform some real research on fault injection, and real-time security.
  • Partial implementation of SecurityManager interface.
  • SSLIOP performance enhancements.
• October 2001
  • Request authentication via Kerberos, as detailed in the CORBA Security Service specification. This requires the Common Secure Interoperability protocol (CSIv2), but initial experiments will be performed via TAO's existing SSLIOP pluggable protocol.
  • Partial implementation of the SecurityLevel2 interfaces. This includes the Credential and Current objects.
  • Partial implementation of the SecurityAdmin interfaces.
  • Begin implementation of SecurityAudit objects. As a side benefit, this project will allow us to research how to secure existing services such as TAO's Real-Time Event Service.
  • Begin implementation of the interfaces in the SecurityReplaceable IDL module. They provide the ability to replace key security components in the ORB with another implementation with ease. Thus, the security components in the ORB become highly extensible.
• November 2001
  The following are slated for post TAO 1.2 betas.
  • Complete implementation of the Commone Secure Interoperability protocol (CSIv2). Needed for more advanced features, such as delegation.
  • Begin implementation of a DomainManager for TAO. This is needed to support security domains.
• December 2001
  • Begin work on the CORBA Non-Repudiation Service.

TAO's Scheduling Service

Currently Implemented Features:

• The scheduling service can be built to use either a null implementation or a strategized implementation of the configuration scheduler.
• The null scheduler implementation, which is built by default, allows the configuration scheduler to be used with applications that require a scheduling service interface, but do not (at least in the current stage of their development, in certain configurations, etc.) make use of the real-time scheduling features it provides.
• The strategized scheduler implementation can be built by #defining TAO_USES_STRATEGY_SCHEDULER, and the appropriate scheduling strategy macro (TAO_USES_RMS_SCHEDULING, TAO_USES_EDF_SCHEDULING, TAO_USES_MUF_SCHEDULING, or TAO_USES_MUF_SCHEDULING) in $ACE_ROOT/ace/config.h. This allows the configuration scheduler to be used with applications that require a specific scheduling strategy. Each scheduling strategy will produce a set of static scheduling priorities, which it will assign to operations based on their RT_Infos. For each static priority, a strategy will also determine the run-time (dynamic) scheduling strategy to use for that priority level.

• Implement heap-based dispatching queues.
• Add support for additional configurability, especially in the type of dispatching strategy (list vs. heap) that will be used to dispatch operations at a given static priority level.
• Benchmark the various alternative strategies to obtain performance profiles across different operation loads and OS platforms.
• Add increased functionality. Requests and suggestions are welcome.

TAO's Logging Service

Current status (Jan 23 2000):

• The basic logging service has been implemented. It can log basic messages from multiple clients. The log scheduling is currently being tested.

• Add persistence and then to add event notification

TAO's support for FT services

Point of contact: Balachandran Natarajan

Current Status:

TAO supports the ORB level requirements to achieve Fault Tolerance for CORBA Objects. The details of the ORB level support is described in the FT spec . Specifically TAO implements the sections 27.2.2, 27.2.3, 27.2.6 thru 27.2.8 of the document.

Future Work:

• Implement 27.2.9 of the document.

TAO's Load Balancer

Point of contact: Ossama Othman

Current Status:

TAO's Load Balancer currently implements the following load balancing algorithms:

• Round Robin
  • Simply forwards requests to the next registered replica in a circular list.
• Minimum Dispersion
  • Basically tries to keep the load on each registered replica close to the average. If the load on a given server is much higher than the average load it is deemed too loaded, at which point the load balancer informs the replica of this condition, which in turn causes it to stop accepting requests until its load is close to the average.

Known Issues:

• Factory must be implemented to allow applications to create load balancers at run-time, rather than at application start up time.

Recent Work:

• Replaced use of DSI for forwarding requests in favor of a ServantLocator. The ServantLocator implementation provides a more robust way of forwarding requests. The DSI implementation was a bit of a hack since ForwardRequest exceptions are only supposed to be thrown from ServantManagers, such as ServantLocators.
• All code is now thread-safe and exception-safe.

Future Work:

• Implement a load balancing algorithm strategy factory.
• Implement load balancing via Interceptors.
• Implement cooperative load balancing support.
• Implement additional load balancing strategies:
  • load based on power consumption
  • load based on the number of requests currently serviced (maybe)
  • adaptive load balancing strategy that utilizes load history to attempt to predict future load conditions
• Active load balancing:
  • Actively poll replicas for current load, as opposed to the currently implemented passive load balancing where the load balancer listens for load reports from the replicas.

Test & Performance Tests

Current Status:

The TAO IDL_Cubit test application makes use of the Naming Service and the server holds a TAO_Naming_Server component.Just running server and client is enough to test the application.

The various tests in the tests/POA test the different features of the Portable Object Adapter interface like Explicit Activation, On Demand Activation,etc..

MT_Cubit:

Current status:

The TAO MT_Cubit test application is meant to serve as a starting point for real-time tests on the TAO system. It comprises the following parts:

• Server. The server creates multiple CORBA objects (servants), each with different real-time priorities. This priority is implemented by using real-time thread support provided by the operating system. Thus, requests sent to a high-priority servant are handled by a high-priority real-time thread, and those sent to a lower priority servant are handled by correspondingly lower priority threads.
• Client. The client component binds to the servants, and sends a stream of CORBA requests to the servants. It measures the response time, i.e. the time taken for the request to complete successfully. In particular, it measures the time taken for requests sent to the high priority servant to complete. The volume of lower priority requests is configurable. The client is thus able to measure the performance of the high-priority servant in the presence of competition from several lower-priority servants.

Future work:

• Study the impacts of scheduling & concurrency strategies on performance.
• Evolve into a testbed for discovering sources of performance non-determinism & priority inversion.

Pluggable:

Current status:

The TAO Pluggable test utilizes ACE Timeprobes to time the latency at various points in the ORB, especially that incurred by the Pluggable Protocols implementation. Comparisons can be made not only between different layers of the ORB, but also between different protocols as they become available.

Future work:

• Add options to redirect the output to a file.
• Script or otherwise automate the piping of the output to a spreadsheet.

ORB-related ACE Changes

Recently Completed Work:

• Added special declaration to OS.h for inet_ntoa and other functions because VxWorks doesn't provide full argument prototypes for these library functions.
• The current caching connector behaves properly in the face of a non-blocking connect request. The "fix" is simply to not support non-blocking connects through the cache. When the connect() fails with EWOULDBLOCK, morph the error to -1 and clean up the request.
• Service handlers obtained from the caching connector are now cleaned up. The application needs to be able to signal that it's not using it any longer, and, when the application encounters an error, needs to effectively close down that connection for good so that a new connection can be initiated.

Added the ability for a Svc_Handler to recycle itself. idle() can be called when the Svc_Handler is done serving a particular connection and can how be recycled. The Svc_Handler now also has a pointer to a recycler that is responsible for managing the connections. The recycler is usually a Cached_Connector.
Added new class ACE_Recycling_Strategy. It defines the interface (and default implementation) for specifying a recycling strategy for a Svc_Handler. This strategy acts as a consular to the Svc_Handler, preparing it for the tough times ahead when the Svc_Handler will be recycled.
Added new class ACE_NOOP_Concurrency_Strategy. It implements a no-op activation strategy in order to avoid calling open on a recycled svc_handler multiple times.
ACE_Cached_Connect_Strategy now implements the ACE_Connection_Recycling_Strategy interface. This allows Svc_Handlers to cache themselves with ACE_Cached_Connect_Strategy when they become idle. It also allows them to purge themselves from the connection cache when the Svc_Handlers close down.
Also added ~ACE_Cached_Connect_Strategy that will cleanup up the connection cache.

None currently scheduled.

The DOVE Demo

DOVE is documented in detail online. This discussion focuses on the following goals:

• Have a DOVE Browser running using Java Beans as vizualization components.
• Have the Event Channel as DOVE Agent running with an Event Consumer in the DOVE Browser.
• Having a DOVE Management Information Base (MIB), which dumps all events transfered on the Event Channel into a file on persistent storage for later reuse.

We have three major components: Observables (monitored metrics), Observers (a Java Bean for displaying the metric) and a DataHandler (for demultiplexing the monitored metrics to the appropriate Observables). Each component inherits from a base class, so that a certain behavior of the components can be assured for each component. Relationships between components are based on these base classes.

The used Java Beans are required to conform to some standards, as they have to support a function called "getProperty" which allows the DOVE Browser to determine if the vizualization capabilities of a specific Java Bean are sufficient to display the metric. A JavaBean is for example a Java Panel which shows a Graph of the delivered doubles. So all metrics can be displayed by this visualization component which can be expressed by a single double.

The DataHandler is connected to the Event Push Consumer (PUSH, because we use the push concept of the Event Service). The Event Push Consumer does not know what kind of data is transported. The only component knowing all the details about the dependencies of the metrics is the DataHandler. This separation allows easy extension and change of the demo.

Object Diagrams are available about this new concept.

Event Service events are used as communication between DOVE Applications and the DOVE Browser. The DOVE MIB analyses the event data field of all events and stores this information into a file. The event data filed is of type CORBA::Any and the DOVE MIB has no notion of what is conveyed in this field. So the DOVE MIB has to discover the content via the embedded type code information. Future work includes:

• Enhancing MIB functionality
• Monitoring the AV Streaming Service

Location Forwarding

For more information see Location forwarding

Global Resources and Leader-Follower Model

For more information see Leader-follower model

Implementation of locate request

For more information see Locate request

Asynchronous Method Invocation

Status:

We've implemented the callback model of the Messaging specification. To activate the AMI for TAO and the TAO IDL compiler define TAO_HAS_CORBA_MESSAGING, TAO_HAS_AMI_CALLBACK in your config.h file. The TAO IDL compiler can generate the AMI stubs, ReplyHandler und reply stubs using the -GC switch.

For an example see $TAO_ROOT\tests\AMI and $TAO_ROOT\examples\AMI.

Finished work:

• Redesign of the IDL compiler to make an addtional pass over the AbstractSyntaxTree and generate the implied-IDL code in memory. This reduced the amount of AMI specific IDL compiler code dramatically.
• Support for exceptions
• Support for attributes
• Support for buffering and batching AMI calls. See Buffered AMI example for details.
• Support for deferred synchronous invocations. Jeff Parsons
• Support for timeouts in combination with AMI calls, response handler gets CORBA::TIMEOUT exception on timeout

Future Work:

• Implementation of the poller model.

Portable Interceptors

For more information see Portable Interceptors

Local Interfaces

Point of contact: Nanbor Wang.

Local interfaces are first defined in the CORBA Component Model specification. For more information on using the local interfaces, please refers to Section 11.1.1 to 11.1.4 of the spec and our short guideline on implementing local objects.

CORBA Standards Conformance

• The full OBV (valueboxes, valuetypes in all sorts of things, value graphs, Anys and value types, abstract interfaces, etc. etc.)
• The IfR database cannot be populated (so it is less than fully operational ;-))
• Fixed data types (who cares)
• GIOP fragments (1.1 and 1.2) are not completely tested.
• The old ServantManager interfaces (POA stuff)
• Domain Managers (useful for security, but otherwise nobody seems to care)
• I think our interpretation of codesets is compliant (we only support one codeset), but would have to check.

Future Work (aka. known problems):

• Supposedly, any constructed types that contains local types become local automatically. TAO_IDL currently doesn't handle the array type very well if one is defined outside the scope of a local interface.
• Need to test local object support more systematically and comprehensively. (Does TAO throw a MARSHAL exception when trying to marshal a local type?)

Back to the TAO documentation index.