Schmidt, D. C., Evaluating Architectures for Multithreaded CORBA Object Request Brokers, CACM Vol. 41, No. 10, 1998  Home/Search   Document Details and Download   Summary   Related Articles   Check

....using thread pools, load balancing, and other techniques on the VirtualRepository to limit the load placed on specific objects. These techniques are well studied and could be implemented here either natively in PIE or by using a transaction platform available in commercial CORBA systems. Schmidt [Sch01] conducted a study of such threading techniques in CORBA serves. Handling persistent queries at PersonalCollection objects is a much easier task in terms of the resource requirements placed on the infrastructure. Most importantly, the PIE model is inherently distributed, so individual PIE servers ....

Douglas C. Schmidt. Evaluating Architectures for Multi-threaded CORBA Object Request Brokers. Communications of the ACM, To Appear., 2001.

....is often used for locating the object implementation and for delivering the request. Daemons add processing overhead and may become performance bottlenecks in highly loaded environments [Sch97, Gok97] The scheduling of threads and processes at the server node may also cause performance problems [Sch98a]. Network bandwidth can be a limiting factor in some applications, such as those running in wireless mobile environments [OMG98c] and those transferring large amounts of data over the network [Gok98a] Network latency is also a well known issue, but its effect on the application behavior can be ....

....is seldom available for detailed analysis. The implementation s threading policies, activation policies, scheduling disciplines, and other operational choices may strongly influence the performance of the system especially with a large number of concurrent clients and intense workloads [Sch98a, Lit98]. Unfortunately, such effects may be difficult to predict only by reading the product s documentation. To cope with this problem, we suggest a two phase testing methodology. First, an initial set of experiments should be designed and carried out for gaining an understanding of the system and the ....

Schmidt, D., Evaluating Architectures for Multi-threaded CORBA Object Request Brokers, Communications of the ACM, Vol. 41, No. 10, October 1998, 54-60.

....ces sources de surcharge ont et er eduites, en plus de celles que nous avons d ej a mentionn ees. Les optimisations qui ont et e faites sontauniveau de la strat egie d invocation des m ethodes, du d emultiplexage des requ etes, de la communication entre ORB [33] et des architectures multithread [63]. ########## ############ ### ######### L optimisation e ectu ee a ce niveau est la diminution de la latence. Pour y parvenir, on a fait une utilisation optimale de la m emoire, r eduit les appels de fonction intra ORB, r eutilis elesrequ etes DII et minimis e l encodage ainsi que le copiage de ....

....en couche [33] ### ############# ########### Plusieurs architectures multithread peuvent etre utilis ees dans les ORB pour am eliorer la performance. L Universit e de Washington a evalu e plusieurs de ces architectures a n de trouver celle qui est la mieux appropri ee pour le temps r eel [63]. Elle a conclu que les architectures thread per request, threadper connexion et thread per servant utilis ees dans la plupart des ORB conventionnels ne sont pas appropri ees pour le temps r eel a cause du non d eterminisme et du mauvais balancementdela charge qu elles impliquent. Elle a aussi ....

D. C. Schmidt. Evaluating Architectures for Multi-threaded CORBA Object Request Brokers. Washington University, Department of Computer Science, 1998.

....4) SDUDOOHO H[HFXWLRQ IORZ ORB A Object B Figure 4 The callback interaction pattern Simultaneous resource possession constructs (Figure 2) are also an appropriate means for the representation of most common threading policies. The most important of them have been experimentally studied in [20]. Thread pool policies, with a fixed number of threads, can be easily represented by selecting the number of tokens, in the modeling construct of Figure 2, to be equal to the number of available threads. Dynamic thread creation and destruction policies can be also easily represented, using the ....

D. C. Schmidt, "Evaluating architectures for multithreaded CORBA object request brokers", Communications of the ACM, vol. 41 (10), pp. 54-60, 1998.

....Dual Intel Pentium III 800 MHz, 512 MB RAM Depending on the specific broker implementation, the parallelism benchmarks need to be run several times or extended to cover all threading strategies. Also, the distribution of the system capacity among the individual threads needs to be evaluated [19]. 6 2.8 Miscellaneous Benchmarks A broker also provides functions that are not directly related to invocation. Typical examples of these are insertion and extraction of data into and from anys, string allocation, or dynamic any inspection. Straightforward benchmarks of these functions can be ....

Schmidt, D.C., Evaluating Architectures for Multi-threaded CORBA Object Request Brokers, CACM Vol. 41, No. 10, 1998

....Dual Intel Pentium III 800 MHz, 512 MB RAM 6 Depending on the specific broker implementation, the parallelism benchmarks need to be run several times or extended to cover all threading strategies. Also, the distribution of the system capacity among the individual threads need to be evaluated [17]. For users: The simplification stems from the fact that a benchmark where a large number of clients issue invocations is difficult to execute. In a local setup, the load generated by the clients interferes with the server. In a network setup, the network bottlenecks can do the same. Also, the ....

Schmidt, D. C., Evaluating Architectures for Multithreaded CORBA Object Request Brokers, CACM Vol. 41, No. 10, 1998

....like worker thread pool or leader follower thread pool architecture. All but the blocking server and the thread per server process incoming requests in parallel. Usually the leader follower thread pool or thread per request architectures are deployed as they are relatively easy to implement [21] and offer acceptable performances [22] The thread pool architecture concurrency is limited by the number of threads in pool. The thread per request ultimately limits the number of clients a server can support to the operating system upper thread limit. When evaluating performance the ....

Douglas C. Schmidt, Evaluating Architectures for Multi-threaded CORBA Object Request Brokers, ACM Special Issue on CORBA

....like worker thread pool or leader follower thread pool architecture. All but the blocking server and the thread perserver process incoming requests in parallel. Usually the leader follower thread pool or thread per request architectures are deployed as they are relatively easy to implement [19] and offer acceptable performances [20] The thread pool architecture concurrency is limited by the number of threads in pool. The thread per request ultimately limits the number of clients a server can support to the operating system upper thread limit. When evaluating performance the ....

Douglas C. Schmidt, Evaluating Architectures for Multi-threaded CORBA Object Request Brokers, ACM Special Issue on CORBA

No context found.

Douglas C. Schmidt, "Evaluating Architectures for Multi-threaded CORBA Object Request Brokers," Communications of the ACM Special Issue on CORBA, vol. 41, no. 10, Oct. 1998.

....ordinary source clients, middle tier servers must communicate with multiple source clients and sink servers. They must therefore be highly scalable to avoid becoming a bottleneck. A common way to improve the throughput of a middle tier server is to multi thread it using various concurrency models [4], such as thread pool, thread per connection, or thread per request. In these models, threads can process new incoming client requests even while other threads are blocked waiting to receive a response from a sink server. Due to the cost of thread creation, context switching, synchronization, and ....

D. C. Schmidt, "Evaluating Architectures for Multi-threaded CORBA Object Request Brokers," Communications of the ACM special issue on CORBA, vol. 41, Oct. 1998.

.... The following are specific known uses of this pattern: CORBA ORBs: The Active Object pattern has been used to implement concurrent ORB middleware frameworks, such as CORBA [6] and DCOM [15] For instance, the TAO ORB [16] implements the Active Object pattern for its default concurrency model [17]. In this design, CORBA stubs correspond to the Active Object pattern s Proxies, which transform remote operation invocations into CORBA Requests. The TAO ORB Core s Reactor is the Scheduler and the socket queues in the ORB Core correspond to the Activation Queues. Developers create Servants that ....

D. C. Schmidt, "Evaluating Architectures for Multi-threaded CORBA Object Request Brokers," Communications of the ACM special issue on CORBA, vol. 41, Oct. 1998.

.... II 2 1 C 3 3 0 [P [P ] P ] S 3 3 2 [P ] S 2 I O SUBSYSTEM Server 1 RUNTIME SCHEDULER [P S 0 S Object Adapter Servants Client [P] Requests Priority [P ] 1 ORB Core Figure 2: ORB Endsystem Priority Inversion Test Configuration a priority based concurrency architecture [31], which is often used by real time applications with deterministic QoS requirements. For instance, avionics mission computing systems [7] commonly execute fixed priority threads corresponding to the rates, e.g. 20 Hz, 10 Hz, 5 Hz, and 1 Hz, at which operations are called by clients. Each client ....

D. C. Schmidt, "Evaluating Architectures for Multi-threaded CORBAObject Request Brokers," Communications of the ACM special issue on CORBA, vol. 41, Oct. 1998.

....Priority based Connection and Concurrency Architectures thread maintains a Connector [21] in thread specific storage. Each Connector manages a map of pre established 2 TAO s ORB Core also can be configured to support thread pool, threadper connection, and single threaded reactive dispatching [17]. 3 connections to servers. A separate connection is maintained for each thread priority in the server ORB. This design enables clients to preserve end to end priorities as requests traverse through ORB endsystems and communication links [11] Figure 4 also shows how the Reactor in each thread ....

....Core. This design vertically integrates TAO s priority based concurrency architecture throughout the ORB endsystem. 4 Periodic threads must specify both a period P and a per period computation time T. Implementing Schedule driven protocol processing in RIO: The RIO subsystem uses a thread pool [17] concurrency model to implement its schedule driven kthreads. This model creates a pool of protocol processing kthreads, known as RIO kthreads, during I O subsystem initialization. Initially these threads are not bound to any connection and therefore are inactive until needed. In general, thread ....

D. C. Schmidt, "Evaluating Architectures for Multi-threaded CORBA Object Request Brokers," Communications of the ACM special issue on CORBA, vol. 41, Oct. 1998.

.... 6 Related Work We have used TAO to research many dimensions of highperformance and real time ORB endsystems, including static [2] and dynamic [25] scheduling, request demultiplexing [7] dispatching [34] and event processing [22] ORB 20 Core connection [32] and concurrency architectures [35], IDL compiler stub skeleton optimizations for synchronous [8] and asynchronous [36] communication, I O subsystem integration [20] evaluation Real time CORBA [11] features [19] fault tolerance features [14, 15] reflective QoS techniques the CORBA Component Model [37] multimedia streaming ....

D. C. Schmidt, "Evaluating Architectures for Multi-threaded CORBA Object Request Brokers," Communications of the ACM special issue on CORBA, vol. 41, Oct. 1998.

.... plied these optimization principle patterns in TAO to address the following ORB design and implementation challenges: Optimizing the server side ORB concurrency model: The concurrency model used to multi thread an ORB has a substantial impact on its performance, predictability, and scalability [24]. However, concurrency models supported in conventional ORBs, such as thread per request or queue based worker thread pools, incur excessive context switching, synchronization, and data movement overhead [21] Therefore, TAO employs a leader followers thread pool model described in Section 2.1. ....

....and data copies, and minimizing GIOP IIOP protocol overhead. Additional optimizations for ORB Core connection management are described in [21] 2. 1 ORB Core Concurrency Model Optimizations Motivation: A common concurrency model used in conventional ORBs is to use a queue based worker thread pool [24]. As shown in Figure 3, the components in this model include a designated I O thread, a request queue, and a pool of worker threads. The I O thread selects(1) on the socket endpoints, 2) reads new client requests, and (3) inserts them into the tail of the request queue. Aworker thread in the pool ....

D. C. Schmidt, "Evaluating Architectures for Multi-threaded CORBA Object Request Brokers," Communications of the ACM special issue on CORBA, vol. 41, Oct. 1998.

....and efficient endpoint demultiplexing and event handler dispatching. 3 The forthcoming real time CORBA specification [41] will support this capability. 4 In addition, TAO s ORB Core can be configured to support thread pool, thread per connection, and single threaded reactive dispatching [42]. 9 When playing the role of a server, TAO s Reactor(s) demultiplex incoming client requests to connection handlers that perform GIOP processing. These handlers collaborate with TAO s Object Adapter to dispatch requests to application level servant operations. Operations can either execute at ....

.... does not allow applications to specify concurrency models, such as thread per request or thread pools, which makes it inflexible for certain types of applications [52] To meet application QoS requirements, it is important to develop ORBs that manage concurrent processing efficiently [42]. Concurrency allows long running operations to execute simultaneously without impeding the progress of other operations. Likewise, preemptive multi threading is crucial to minimize the dispatch latency of real time systems [10] Concurrency is often implemented via the multi threading ....

[Article contains additional citation context not shown here]

D. C. Schmidt, "Evaluating Architectures for Multi-threaded CORBAObject Request Brokers," Communications of the ACM special issue on CORBA, vol. 41, Oct. 1998.

.... principle patterns to address the following TAO Object Adapter and ORB Core design and implementation challenges: Optimizing the server side ORB concurrency model: The concurrency model used to multi thread an ORB has a substantial impact on its performance, predictability, and scalability [20]. However, concurrency models supported in conventional ORBs, such as thread per request or queue based worker thread pools, incur excessive context switching, synchronization, and data movement overhead. Therefore, TAO employs a leader followers based thread pool model that requires no heap ....

....and data copies, and minimizing GIOP IIOP protocol overhead. Additional optimizations for ORB Core connection management are described in [16] 2. 1 ORB Core Concurrency Model Optimizations Motivation: A common concurrency model used in conventional ORBs is to use a queue based worker thread pool [20]. As shown in Figure 3, the components in this model include a SERVANTS SERVANTS ORB ORB CORECORE 1: select( 1: select( I I OO SUBSYSTEMSUBSYSTEM 5: dispatch upcall( 5: dispatch upcall( 2: read( 2: read( 3: enqueue( 3: enqueue( 4: dequeue( 4: dequeue( I I OO THREAD THREAD WORKER ....

D. C. Schmidt, "Evaluating Architectures for Multi-threaded CORBA Object Request Brokers," Communications of the ACM special issue on CORBA, vol. 41, Oct. 1998.

No context found.

D. C. Schmidt, "Evaluating Architectures for Multi-threaded CORBA Object Request Brokers," Communications of the ACM special issue on CORBA, vol. 41, Oct. 1998.

....Core that received the operation. 3 The forthcoming real time CORBA specification [41] will support this capability. 4 In addition, TAO s ORB Core can be configured to support other concurrency architectures, including thread pool, thread per connection, and singlethreaded reactive dispatching [42]. The server sets priority model is well suited for deterministic real time applications since it minimizes priority inversion and non determinism in TAO s ORB Core [45] In addition, it reduces context switching and synchronization overhead since servant state must be locked only if servants ....

D. C. Schmidt, "Evaluating Architectures for Multi-threaded CORBA Object Request Brokers," Communications of the ACM special issue on CORBA, vol. 41, Oct. 1998.

....ORB Cores that use connection multiplexing incur priority inversions when low priority threads awaiting replies from a server block out higher priority threads awaiting replies from the same server. ffl Thread pools: On the server side, ORB Core concurrency architectures often use thread pools [26] to select a thread in which to process an incoming request. However, conventional ORBs do not provide programming interfaces that allow real time applications to assign the priority of threads in this pool. Therefore, the priority of a thread in the pool is often inappropriate for the priority of ....

....computing systems [22] which possess a small, fixed number of connections and priority levels, where each priority level maps to an OS thread priority. 3. 2 Alternative ORB Core Concurrency Architectures There are a variety of strategies for structuring the concurrency architecture in an ORB [26]. Below, we describe a number of alternative ORB Core concurrency architectures, focusing on server side concurrency. 3.2.1 The Worker Thread Pool Architecture ffl Overview: This ORB concurrency architecture uses a design similar to the active connection architecture described in Section 3.1.1. ....

D. C. Schmidt, "Evaluating Architectures for Multi-threaded CORBA Object Request Brokers," Communications of the ACM special issue on CORBA, vol. 41, Oct. 1998.

....and event handler dispatching. 3 The forthcoming real time CORBA specification [41] will support this capability. 4 In addition, TAO s ORB Core can be configured to support other concurrency architectures, including thread pool, thread per connection, and singlethreaded reactive dispatching [42]. When playing the role of a server, TAO s Reactor(s) demultiplex incoming client requests to connection handlers that perform GIOP processing. These handlers collaborate with TAO s Object Adapter to dispatch requests to application level servant operations. Operations can either execute with ....

....computing systems [17] which possess a small, fixed number of connections and priority levels, where each priority level maps to an OS thread priority. 5. 2 Alternative ORB Core Concurrency Architectures There are a variety of strategies for structuring the multithreading architecture in an ORB [42]. Below, we describe a number of alternative ORB Core multi threading architectures, focusing on server side multi threading. 5.2.1 The Worker Thread Pool Architecture ffl Overview: This ORB multi threading architecture uses a design similar to the active connection architecture described in ....

[Article contains additional citation context not shown here]

D. C. Schmidt, "Evaluating Architectures for Multi-threaded CORBA Object Request Brokers," Communications of the ACM special issue on CORBA, vol. 41, Oct. 1998.

....more CORBA implementations. Each architecture is evaluated in terms of its ability to support the aggregate processing capacity of ORB endsystem components and application operations in one or more threads. There are a variety of strategies for structuring the multithreading architecture in an ORB [8]. Below, we describe a number of alternative ORB Core multi threading architectures, focusing on server side multi threading. The Thread per Request Architecture ffl Overview: The thread per request architecture [9] handles each request from a client in a separate thread of control. As shown in ....

D. C. Schmidt, "Evaluating Architectures for Multi-threaded CORBA Object Request Brokers," Communications of the ACM special issue on CORBA, vol. 41, Oct. 1998.

No context found.

Schmidt, D. C., Evaluating Architectures for Multithreaded CORBA Object Request Brokers, CACM Vol. 41, No. 10, 1998

No context found.

Douglas C. Schmidt. Evaluating Architecture for Multi-threaded CORBA Object Request Brokers, ACM Special Issue on CORBA, Krishnan Seetharaman, October 1998, Vol. 41, No. 10.

No context found.

Douglas C. Schmidt, Evaluating Architectures for Multi-threaded CORBA Object Request Brokers, submitted to the CACM special issue on CORBA edited by Krishnan Seetharaman

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC