Distributed C++

Abstract

Distributed C++ (DC++) is vanilla C++ with additional constructs added to support concurrency. It currently runs on homoogeneous workstations (HP Series 9000, Model 300/400/800/700) using BSD sockets or Mach ports for communication.The new constructs are domains and lhreads. Domains are "spatial/temporM" entities. Each domain is a logically separate (or encapsulated) address space. All data passed or returned between domains is copied (all data involved in intradomain calls follow normal C++ conventions). Concurrency is achieved by creating multiple threads of control. Only one thread of control may be active in a given domain at any time. Other threads attempting entry into an occupied domain are placed on a FIFO queue outside the domain for later entry. This model of programming adapts quite well to distributed memory parallel processors since domains reside on specific processors (although there may be more than one domain multitasking on a single processor).A domain is a first class construct in DC++. It is constructed by specifying a processor number. DcDomainGid doms [MAX_PROCESSORS] ; for (int p = O; p < MAX_PKOCESSORS; p++) doms[p] = DcMakeDomain(p);The "handle" returned by DcMakeDomain uniquely indicates a specific domain on a particular processor.Operator new is overloaded such that objects may either be created locally or remotely in specific domains.rayptr* local_ray = new rayptr();" rayptr* scanline_rays [YMAX] ; for (int y = O; y < YMAX; y++) scanline_rays[y] = new rayptr(doms[y Y, MAX_P ROCESSORS] ) ;Once the object has been created the program does not need to concern itself with whether or not a call is local or remote (the call syntax is identical). local_ray->fire_ray () ; for (y = O; y < YMAX; y++) scanline_rays[y]->fire_ray();This is accomplished via the "handle/body" class idiom. The programmer writes the body class while the DC++ compiler generates a handle class for the body. The handle class does either a local call or transparently sets up a remote procedure call.Concurrency is achieved by explicitly creating threads.DcDomainGid ping_domain = DcMakeDomain(O); DcDomainGid ponE_domain = DcMakeDomain(1) ; DcMakeThread(ping_pong, pong_domain, ping_domain, False) ;The arguments to DcMakeThread are a procedure to run, an argument(s) to the procedure, the domain in which to apply the procedure, and a flag to specify if a return value is expected. If a return value is expected then a placeholder is returned so the thread which is creating the new thread may continue execution. Later, when the value is needed, an explicit touch is used to ensure that the return value is ready. If not, the thread doing the touch will block until the value has been computed. retval = DcMakeThread(ping_pong, pong_domain, pinE_domain , True); ooo a = b + DcTouch(retval);Domains and threads are the fundamental constructs added to C++ to provide multiple address spaces and threads of control. Experience has shown that additional constructs are useful in many distributed applications. Besides the above "transparent" rem...

A survey of concurrent object-oriented languages

“…Murat Karaorman −→ murat@cs.ucsb.edu John Bruno −→ bruno@cs.ucsb.edu Concurrent Aggregates, CA [106][107][108][109][110] ¡ ¡ À À ¨¡ ¡ À À ConcurrentSmalltalk [111] ¡ ¡ À À ¨¡ ¡ À À [113][114][115] ¡ ¡ À À ¨¡ ¡ À À [122,123] ¡ ¡ À À ¨¡ ¡ À À Distributed C++, DC++ [125,126] ¡ ¡ À À ¨¡ ¡ À À Distributed Eiffel [127] ¡ ¡ À À ¨¡ ¡ À…”

PEPt – A Minimal RPC Architecture

“…The specification and implementation of Remote Procedure Call (RPC) [1] systems such as DCE [2], distributed versions of C++ [3] [4], COM/DCOM [5], CORBA [6], RMI [7], RMI-IIOP [8], XML-RPC [9] and SOAP [10], seems to traverse the same ground repeatedly. One way to avoid reinventing the wheel is to isolate the basic building blocks.…”