R. Gopalakrishnan and G. M. Parulkar. Bringing real-time scheduling theory and practice closer for multimedia computing. In ACM SIGMETRICS Conference (Philadelphia, PA), pages 1--12, May 1996. 13  Home/Search   Document Details and Download   Summary   Related Articles   Check

....since it is similar to the CPU speeds on many DRE platforms, such as avionics mission computing [24, 25] 12 are met. An important metric for evaluating the performance of a real time system is its schedulable bound, which is the maximum resource utilization possible without missing deadlines [35]. The schedulable bound of TAO s Real time Event Service is therefore the maximum CPU utilization that suppliers and consumers can achieve without missing deadlines. With rate monotonic scheduling, higher rate tasks are supposed to preempt lower rate tasks. For TAO s Real time Scheduling Service ....

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing," in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

....partitioned [31] This is particularly true for preemption caused by external events such as network interrupts. One can account for the cache miss penalty due to preemption via careful schedulability analysis [32] but frequent preemption still degrades available CPU capacity, as also observed in [33]. Moreover, an important implication of arbitrary preemption for the proposed architecture is that a handler may 23 get preempted just before initiating transmission, even though it had finished preparing a packet for transmission, thus idling the link. Cooperative preemption, on the other hand, ....

....the communication subsystem. Modeling of real time operating systems: Since the analysis presented in Section 3. 2 is geared towards the real time communication needs of distributed systems, it compliments recent efforts to bridge the gap between the theory and practice for real time systems [33, 71 73]. The implications of priority inversion due to non preemptible critical sections was studied in [74] however, preemption costs (context switches and cache misses) and the resulting degradation in useful resource capacity were not considered. OS support for QoS sensitive computation and ....

R. Gopalakrishnan and G. M. Parulkar, "Bringing real-time scheduling theory and practice closer for multimedia computing," in Proc. of ACM SIGMETRICS, pp. 1--12, May 1996.

....the issues and tradeoffs involved in realizing QoS sensitive communication subsystems are not considered. Modeling of multimedia real time operating systems: A number of recent efforts have attempted to bridge the gap between theory and practice for real time systems and multimedia computing [29, 69, 95, 98]. The admission control extensions developed in Chapter 4 are geared towards the real time communication needs of distributed systems, and hence complement these efforts. The implications of priority inversion due to non preemptible critical sections was studied in [121] however, preemption costs ....

....[110] This is particularly true for preemption caused by external events such as network interrupts. One can account for the cache miss penalty due to preemption via careful schedulability analysis [105] but frequent preemption still degrades available CPU capacity, as also observed in [69]. Moreover, an important implication of arbitrary preemption for the proposed architecture is that a handler may get preempted just before initiating transmission, even though it had finished preparing a packet for transmission, thus idling the link. Cooperative preemption, on the other hand, ....

[Article contains additional citation context not shown here]

R. Gopalakrishnan and G. M. Parulkar, "Bringing real-time scheduling theory and practice closer for multimedia computing," in Proc. of ACM SIGMETRICS, pp. 1--12, May 1996. 225

....and a GIOP connection handler to service that queue. TAO s I O subsystem uses the port number contained in arriving requests as a demultiplexing key to associate requests with the appropriate socket queue. This design minimizes priority inversion through the ORB endsystem via early demultiplexing [22, 23, 14], which associates requests arriving on network interfaces with the appropriate real time thread that services the target servant. As described in Section 3.3.1, early demultiplexing is used in TAO to vertically integrate the ORB endsystem s QoS support from the network interface up to the ....

....integrating packets received at the network interface with the corresponding thread priorities in TAO s ORB Core. Schedule driven protocol processing: This feature performs all protocol processing in the context of kernel threads that are scheduled with the appropriate real time priorities [22, 23, 36, 25]. Our design schedules network interface bandwidth and CPU time to minimize priority inversion and decrease interrupt overhead during protocol processing. Dedicated STREAMS: This feature isolates request packets belonging to different priority groups to minimize FIFO queueing and shared resource ....

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing, " in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

....ORB Core. Section 3.1 describes how the TAO s RIO subsystem implements early demultiplexing. Schedule driven protocol processing: To minimize threadbased priority inversions, this feature performs all protocol processing with threads that are scheduled with the appropriate real time priorities [32, 31, 14]. RIO s design schedules network interface bandwidth and CPU time to minimize priority inversion and decrease interrupt overhead during protocol processing. Section 3.2 describes how the TAO s RIO subsystem implements schedule driven protocol processing. Dedicated STREAMS: This feature addresses ....

....RIO kthreads typically service their associated RIO queues at the periodic rate specified by an application. In addition, RIO can allocate kthreads to process the output RIO queue. 3. 2 Schedule driven Protocol Processing Context: Many real time applications require periodic I O processing [32]. For example, avionics mission computers must process sensor data periodically to maintain accurate situational awareness [23] If the mission computing system fails unexpectedly, corrective action must occur immediately. Problem: Protocol processing of input packets in Solaris STREAMS is ....

[Article contains additional citation context not shown here]

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing," in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

....to distributed applications. # Capturing and documenting the key design patterns [25] necessary to develop, maintain, configure, and extend real time ORB endsystems. In addition to providing a real time ORB, TAO is an integrated ORB endsystem that consists of a high performance I O subsystem [27, 28] and an ATM Port Interconnect Controller (APIC) 29] Figure 4 illustrates the main components in TAO s ORB endsystem architecture. 2.3.2 Requirements for High performance and Real time ORB Endsystems The remainder of this section describes the requirements and features of ORB endsystems ....

....and a GIOP connection handler to service that queue. TAO s I O subsystem uses the port number contained in arriving requests as a demultiplexing key to associate requests with the appropriate socket queue. This design minimizes priority inversion through the ORB endsystem via early demultiplexing [27, 28, 29], which associates requests arriving on network interfaces with the appropriate real time thread that services the target servant. As described in Section 8, early demultiplexing is used in TAO to vertically integrate the ORB endsystem s QoS support from the network interface up to the application ....

[Article contains additional citation context not shown here]

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing, " in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

....to build high performance, real time ORBs. We are transferring this technology to ORB vendors who will productize and commercialize our research results. In addition to providing a real time ORB, TAO is an integrated ORB endsystem architecture that consists of a highperformance I O subsystem [10, 11] and an ATM Port Interconnect Controller (APIC) 12] This ORB endsystem architecture is designed to support end to end gigabit data rates and 10 msec latency to CORBA applications. In addition, TAO is designed with a layered architecture, which can also run on embedded platforms that are linked ....

....such as CPUs, memory, storage throughput, network adapter throughput, and network connection bandwidth and latency. For instance, OS scheduling mechanisms must allow high priority client requests to run to completion and prevent them from being blocked indefinitely by lower priority requests [11, 15]. Section 3.1 describes the OS I O subsystem and network adapter integrated with TAO. This infrastructure is designed to support end to end gigabit data rates and 10 msec latency to CORBA applications. In addition, TAO is designed to be layered atop other OS network platforms (such as 1553 buses, ....

[Article contains additional citation context not shown here]

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing, " in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996. 19

....Quality of Service support in operating systems for multimedia applications. A real time upcall facility is used to implement periodic activity in a process [14] a handler routine is called in the user program by the kernel according to their Rate Monotonic with Delayed Preemption (RMDP) policy [15]. The real time upcall is similar to the mechanism in the DQM Architecture in which the DQM communicates with the individual applications. An integrated, end to end approach using real time upcalls has been shown effective for a modified httpd daemon serving video streams [5] This work is very ....

R. Gopalakrishnan and Guru M. Parulkar. Bringing real-time scheduling theory and practice closer for multimedia computing. In Proceedings of the ACM SIGMETRICS Conference on Measurement and Modeling of Computer Systems, Philadelphia, May 1996.

....distributed applications. ffl Capturing and documenting the key design patterns necessary to develop, maintain, and extend real time ORB middleware [21] In addition to providing a real time ORB, TAO is an integrated ORB endsystem architecture that consists of a highperformance I O subsystem [22, 23] and an ATM Port Interconnect Controller (APIC) 20] Figure 4 illustrates the primary components in TAO. Implementing an effective framework for real time CORBA requires ORB developers to address two broad categories of issues: QoS specification and QoS enforcement. First, realtime applications ....

....execution period, and communication class. Communication classes supported by TAO include ISOCHRONOUS (for continuous media) BURST (for bulk data) MESSAGE (for small messages with low delay requirements) and MESSAGE STREAM (for message sequences that must be processed at a certain rate) [23]. In addition to transporting application QoS attributes, TAO s RIOP is designed to map CORBA GIOP on a variety of networks (including high speed networks like ATM LANs and ATM IP WANs [33] and can be customized for specific application requirements. For instance, to support applications that do ....

[Article contains additional citation context not shown here]

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing, " in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

....integrating packets received at the network interface with the corresponding thread priorities in TAO s ORB Core. Schedule driven protocol processing: This feature performs all protocol processing in the context of kernel threads that are scheduled with the appropriate real time priorities [13, 14, 15]. RIO s design schedules network interface bandwidth and CPU time to minimize priority inversion and decrease interrupt overhead during protocol processing. Dedicated STREAMS: This feature isolates request packets belonging to different priority groups to minimize FIFO queueing and shared ....

....RIO kthreads typically service their associated RIO queues at the periodic rate specified by an application. In addition, RIO can allocate kthreads to process the output RIO queue. 2. 2 Schedule driven Protocol Processing Context: Many real time applications require periodic I O processing [13]. For example, avionics mission computers must process sensor data periodically to maintain accurate situational awareness [19] If the mission computing system fails unexpectedly, corrective action must occur immediately. Problem: Protocol processing of input packets in Solaris STREAMS is ....

[Article contains additional citation context not shown here]

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing," in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

.... Request Broker, QoS enabled OO Middleware 1 Introduction Next generation distributed multimedia applications, such as video conferencing, Internet telephony, and video on demand, require endsystems that can provide statistical and deterministic quality of service (QoS) guarantees for latency [1], bandwidth, and reliability [2] The following trends are shaping the evolution of software development techniques for these distributed multimedia applications and endsystems: Increased focus on middleware and integration frameworks: There is a trend in multimedia R D projects away from ....

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing," in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

....integrating packets received at the network interface with the corresponding thread priorities in TAO s ORB Core. Schedule driven protocol processing: This feature performs all protocol processing in the context of kernel threads that are scheduled with the appropriate real time priorities [26, 27, 28, 14]. RIO s design schedules network interface bandwidth and CPU time to minimize priority inversion and decrease interrupt overhead during protocol processing. Dedicated STREAMS: This feature isolates request packets belonging to different priority groups to minimize FIFO queueing and shared resource ....

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing," in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

....to build high performance, real time ORBs. We are transferring this technology to ORB vendors who will productize and commercialize our research results. In addition to providing a real time ORB, TAO is an integrated ORB endsystem architecture that consists of a highperformance I O subsystem [10, 11] and an ATM Port Interconnect Controller (APIC) 12] This ORB endsystem architecture is designed to support end to end gigabit data rates and 10 msec latency to CORBA applications. In addition, TAO is designed to run within embedded OS platforms over various real time interconnects (such as 1553 ....

....such as CPUs, memory, storage throughput, network adapter throughput, and network connection bandwidth and latency. For instance, OS scheduling mechanisms must allow high priority client requests to run to completion and prevent them from being blocked indefinitely by lower priority requests [11, 15]. Section 3.1 describes the OS I O subsystem and network adapter integrated with TAO. This infrastructure is designed to support end to end gigabit data rates and 10 msec latency to CORBA applications. In addition, TAO is designed to be layered atop other OS network platforms (such as 1553 buses, ....

[Article contains additional citation context not shown here]

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing, " in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

....grant NCR 9628218, Siemens, and Sprint. 1 Introduction Next generation distributed real time applications, such as teleconferencing, avionics mission computing, and process control, require endsystems that can provide statistical and deterministic quality of service (QoS) guarantees for latency [1], bandwidth, and reliability [2] The following trends are shaping the evolution of software development techniques for these distributed real time applications and endsystems: Increased focus on middleware and integration frameworks: There is a trend in real time R D projects away from ....

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing, " in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

....implementation where one thread is responsible for dispatching all queued requests. This requires that consumers cooperatively preempt themselves when a higher priority consumer becomes runnable. This model of deferred preemption is based on a Real time Upcall (RTU) concurrency mechanism [32]. The primary benefit of the RTU model is its ability to reduce the context switching, synchronization, and data movement overhead incurred by preemptive multi threading implementations. However, preemption is deferred to the extent that consumers check to see if they must preempt themselves. This ....

....and (2) consumers receive these events in time to meet their deadlines. An important metric for evaluating the performance of the RT Event Service is the schedulable bound. The schedulable bound of a real time schedule is the maximum resource utilization possible without deadlines being missed [32]. Likewise, the schedulable bound of the RT Event Service is the maximum CPU utilization that suppliers and consumers can achieve without missing deadlines. For TAO s Real time Scheduling Service to guarantee the schedulability of a system, i.e. that all tasks meet their deadlines, high priority ....

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing, " in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

....to be notified or blocked at the earliest. The scheduling delay associated with session level shaping can be largely eliminated (or at least made predictable) by employing QoS sensitive CPU scheduling policies. Such fine grain multiplexing of communication threads has been adopted and analyzed in [14] for RM scheduling, and in [15] for EDF scheduling, respectively. The architecture described in [15] decouples protocol processing priority from application priority, deriving the former derived from the traffic and QoS specification on each connection. Accurate traffic shaping is realized via ....

R. Gopalakrishnan and G. M. Parulkar, "Bringing real-time scheduling theory and practice closer for multimedia computing ", in Proc. of ACM SIGMETRICS, pp. 1--12, May 1996.

....to distributed applications. ffl Capturing and documenting the key design patterns [25] necessary to develop, maintain, configure, and extend real time ORB endsystems. In addition to providing a real time ORB, TAO is an integrated ORB endsystem that consists of a high performance I O subsystem [27, 28] and an ATM Port Interconnect Controller (APIC) 29] Figure 4 illustrates the main components in TAO s ORB endsystem architecture. 1.3.2 Requirements for High performance and Real time ORB Endsystems The remainder of this section describes the requirements and features of ORB endsystems ....

....and a GIOP connection handler to service that queue. TAO s I O subsystem uses the port number contained in arriving requests as a demultiplexing key to associate requests with the appropriate socket queue. This design minimizes priority inversion through the ORB endsystem via early demultiplexing [27, 28, 29], which associates requests arriving on network interfaces with the appropriate real time thread that services the target servant. As described in Section , early demultiplexing is used in TAO to vertically integrate the ORB endsystem s QoS support from the network interface up to the ....

[Article contains additional citation context not shown here]

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing, " in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

....of such architectures, their degree of run time predictability must be thoroughly assessed. In this paper, we address the problem of non preemptive scheduling of periodic, real time threads on a uniprocessor MLC architecture. Non preemptive scheduling is a natural choice for many systems [15, 16], since uncontrolled preemption can give rise to a very large number of context switches, hence impeding the schedulability of the application. Moreover, non preemptive scheduling guarantees exclusive access to shared resources and thus eliminates much of the need for synchronization and its ....

....selection mechanism in a multithreaded architecture with multiple on chip register sets. However, their work only evaluated scheduling strategies whose objective is to minimize the average application completion time, and not to meet explicit real time constraints. Gopalakrishnan and Parulkar [16] proposed the rate monotonic with delayed preemption (RMDP) heuristic for systems with uniform context switch cost. Their results showed that the contextswitch overhead can be significantly lowered by using a delayed preemption scheme. The work that comes closest to ours is that by Cheng et al. ....

R. Gopalakrishnan and G. M. Parulkar, "Bringing Real-Time Scheduling Theory and Practice Closer for Multimedia Computing," Proc. of the ACM SIGMETRICS Conference on Measurements and Modeling of Computer Systems, Philadelphia, Pennsylvania, May 23--26, 1996, pp. 1--12.

.... OO Middleware, Performance Measurements 1 Introduction Next generation distributed real time applications, such as teleconferencing, avionics mission computing, and process control, require endsystems that can provide statistical and deterministic quality of service (QoS) guarantees for latency [1], bandwidth, and reliability [2] The following trends are shaping the evolution of software development techniques for these distributed real time applications and endsystems: Increased focus on middleware and integration frameworks: There is a trend in real time R D projects away This work ....

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing, " in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

....grant NCR 9628218, Siemens, and Sprint. 1 Introduction Next generation distributed real time applications, such as teleconferencing, avionics mission computing, and process control, require endsystems that can provide statistical and deterministic quality of service (QoS) guarantees for latency [1], bandwidth, and reliability [2] The following trends are shaping the evolution of software development techniques for these distributed real time applications and endsystems: Increased focus on middleware and integration frameworks: There is a trend in real time R D projects away from ....

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing," in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

.... data copying between the filesystem and networking layer by using a special buffer called an MMBUF that acts as both a network mbuf or a filesystem buffer [12] This system is used to transmit video data from a disk file over an ATM network using real time upcalls to maintain quality of service [29]. This system could be adapted to use a mechanism like UVM s page loanout to move data without copying it (or mapping it) through a user address space. 74 New designs in host network interfaces also look to take advantage of services that a VM system such as UVM can offer. For example, the APIC ....

R. Gopalakrishnan and G. Parulkar. Bringing real-time scheduling theory and practice closer for multimedia computing. In Proceedings ACM SIGMETRICS, May 1996.

....to distributed applications. ffl Capturing and documenting the key design patterns [25] necessary to develop, maintain, configure, and extend real time ORB endsystems. In addition to providing a real time ORB, TAO is an integrated ORB endsystem that consists of a high performance I O subsystem [27, 28] and an ATM Port Interconnect Controller (APIC) 29] Figure 4 illustrates the main components in TAO s ORB endsystem architecture. 2.3.2 Requirements for High performance and Real time ORB Endsystems The remainder of this section describes the requirements and features of ORB endsystems ....

....and a GIOP connection handler to service that queue. TAO s I O subsystem uses the port number contained in arriving requests as a demultiplexing key to associate requests with the appropriate socket queue. This design minimizes priority inversion through the ORB endsystem via early demultiplexing [27, 28, 29], which associates requests arriving on network interfaces with the appropriate real time thread that services the target servant. As described in Section , early demultiplexing is used in TAO to vertically integrate the ORB endsystem s QoS support from the network interface up to the ....

[Article contains additional citation context not shown here]

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing, " in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

.... of the packet into an on going computation ( la Active Messages [15] The scheduling policy used could be a simple Unix like priority scheme, or a periodic real time The APIC Approach to High Performance Network Interface Design Page 20 scheme such as that used with Real Time Upcalls (RTUs) [18,19]. In the latter case, it is possible to entirely disable interrupts for a connection, and the application can periodically (in an RTU) check the receive queue for packet arrivals. This scheme has all the benefits of clocked interrupts [30] and has the advantage that low latency applications can ....

Gopalakrishnan, R., and Parulkar, G.M., "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing," Proc. ACM SIGMETRICS 96.

....integrating packets received at the network interface with the corresponding thread priorities in TAO s ORB Core. Schedule driven protocol processing: This feature performs all protocol processing in the context of kernel threads that are scheduled with the appropriate real time priorities [6, 7, 8, 9]. RIO s design schedules network interface bandwidth and CPU time to minimize priority inversion and decrease interrupt overhead during protocol processing. Dedicated STREAMS: This feature isolates request packets belonging to different priority groups to minimize FIFO queueing and shared ....

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing," in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

....implementation where one thread is responsible for dispatching all queued requests. This requires that consumers cooperatively preempt themselves when a higher priority consumer becomes runnable. This model of deferred preemption is based on a Real time Upcall (RTU) concurrency mechanism [25]. The primary benefit of the RTU model is its ability to reduce the context switching, synchronization, and data movement overhead incurred by preemptive multithreading implementations. However, preemption is delayed to the extent that consumers check to see if they must preempt themselves. This ....

....and (2) consumers receive these events in time to meet their deadlines. An important metric for evaluating the performance of the RT Event Service is the schedulable bound. The schedulable bound of a real time schedule is the maximum resource utilization possible without deadlines being missed [25]. Likewise, the schedulable bound of the RT Event Service is the maximum CPU utilization that supplier and consumers can achieve without missing deadlines. For TAO s Real time Scheduling Service to guarantee the schedulability of a system (i.e. all tasks meet their deadlines) high priority ....

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing," in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

....starving best effort traffic. This suggests that the data exporter threads, i.e. those performing fragmentation and other processing of network data, be scheduled for execution using fine grain preemption. Such fine grain multiplexing of communication threads has been adopted and analyzed in [18, 19] for RM scheduling, and in [20, 21] for EDF scheduling, respectively. The architecture described in [21] decouples protocol processing priority from application priority, deriving the former derived from the traffic and QoS specification on each connection. Accurate traffic shaping is realized via ....

....connections, using policing with TCP over an ATM link yields an unpredictable throughput because packets are sent over different VCs and arrive at the receiving end out of order. Our work complements recent work on QoS sensitive CPU scheduling of applications [14, 15, 16] and protocol processing [18, 19, 20, 21] at end hosts. While these efforts focus on CPU scheduling, our primary focus is on the QoS support architecture exported to sockets based applications. With appropriate CPU scheduling support, our QoS architecture enables new and legacy applications to utilize end to end QoS on communication. We ....

R. Gopalakrishnan and G. M. Parulkar, "Bringing real-time scheduling theory and practice closer for multimedia computing", in Proc. of ACM SIGMETRICS, pp. 1--12, May 1996.

....than using other OS facilities, i.e. threads, to implement a context switch. Therefore, we have not considered this approach in our applications. Synchronous preemption does have potential application in some real time systems. A well developed implementation is real time upcalls (RTUs) [45]. Because it uses deferred preemption, preemption latency is sacrificed. However, context switch overhead is greatly reduced because there is less state to save restore. In addition, preemption only occurs at known points in the application, so that concurrency control does not require locking. ....

....Real time Scheduling Service for system wide scheduling. Therefore, TAO allows applications to use an event driven, multi threaded, or hybrid concurrency models. The Reactor per rate group concurrency model used in TAO is similar to the Real time Upcall (RTU) concurrency mechanism described in [45]. The primary difference is that RTUs use a single thread to dispatch all the operations in a process. Since the basic RTU model is non preemptive, this design is prone to longer duration priority inversions than the Reactor per rate group used by TAO. 8 Concluding Remarks Conventional ....

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing, " in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

....1 Introduction 1. 1 Emerging Trends in Distributed Real time Systems Next generation distributed and real time applications, such as video on demand, teleconferencing, and avionics, require endsystems that can provide statistical and deterministic quality of service (QoS) guarantees for latency [1], bandwidth, and reliability [2] The following trends are shaping the evolution of software development techniques for these distributed realtime applications and endsystems: Increased focus on middleware and integration frameworks: There is a general industry trend away from programming ....

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing," in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

....and a GIOP connection handler to service that queue. TAO s I O subsystem uses the port number contained in arriving requests as a demultiplexing key to associate requests with the appropriate socket queue. This design minimizes priority inversion through the ORB endsystem via early demultiplexing [25, 26, 17], which associates requests arriving on network interfaces with the appropriate real time thread that services the target servant. Early demultiplexing is used in TAO to vertically integrate the ORB endsystem s QoS support from the network interface up to the application servants. 2.2 Handling ....

....integrating packets received at the network interface with the corresponding thread priorities in TAO s ORB Core. Schedule driven protocol processing: This feature performs all protocol processing in the context of kernel threads that are scheduled with the appropriate real time priorities [25, 26, 42, 28]. RIO s design schedules network interface bandwidth and CPU time to minimize priority inversion and decrease interrupt overhead during protocol processing. Dedicated Streams: This feature isolates request packets belonging to different priority groups to minimize FIFO queueing and shared ....

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing, " in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

No context found.

Gopalakrishnan, R., and Parulkar, G.M., "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing," Proc. ACM SIGMETRICS 96. 130

....to specify a flexible bandwidth requirement. The scheduler ensures that at least one, and up to , PDUs will be processed in a period, depending on the system load. 3.2. The RMDP Scheduling Policy A complete analysis of Rate Monotonic with Delayed Preemption (RMDP) scheduling can be found in [17]. We illustrate this policy only briefly to aid understanding of some of our experimental results. TCP IP Implementation with Endsystem QoS 5 Generate PDUs pdu count pdu countB p yield req pdu countb p interrupt yield req = TRUE N Y Y Enqueue PDUs Switch resume USER KERNEL N ....

R. Gopalakrishnan and G.M. Parulkar. Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing. ACM SIGMETRICS, May 1996.

....endsystem qos is based on the Real time Upcall (rtu) mechanism. rtu is a real time scheduling mechanism that allows applications to explicitly register a code segment (an rtu handler) with associated qos parameters. The rtu scheduler uses Rate monotonic with Delayed Preemption (rmdp) policy [17] that exploits the iterative nature of protocol processing to reduce context switching overhead and to increase endsystem efficiency. More details of the rtu mechanism can be found in [1] In this section, we briefly introduce the necessary background information by first examining how the cpu ....

....data units to be processed in one period. Depending on the system load, the scheduler ensures that at least b p , and up to B p , pdus will be processed in each period. 3. 2 The RMDP Scheduling Policy A complete analysis of Rate Monotonic with Delayed Preemption (rmdp) scheduling can be found in [17]. We illustrate this policy only briefly to aid understanding of some of our experimental results. Figure 2 shows the rmdp scheduling algorithm. It is designed to exploit the iterative nature of multimedia processing by treating each iteration as an atomic operation. Preemptions only occur at the ....

R. Gopalakrishnan and G.M. Parulkar. Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing. ACM SIGMETRICS, May 1996.

....such as CPUs, memory, storage throughput, network adapter throughput, and network connection bandwidth and latency. For instance, OS scheduling mechanisms must allow high priority CORBA requests to run to completion and prevent them from being blocked indefinitely by lower priority operations [7, 8]. Section 4.1 describes an OS I O subsystem and network adapter that can provide end to end gigabit data rates and #10 msec latency to CORBA applications. # Optimized real time communication protocols: The throughput, latency, and reliability requirements of multimedia applications like ....

....the layered demultiplexing and dispatching mechanisms in conventional ORBs neither schedule nor prioritize demultiplexing behavior. Therefore, the ORB endsystem must provide mechanisms (such as packet filters [9] de layered protocol stacks [10] direct demultiplexing [11] and real time upcalls [7]) that perform CORBA request demultiplexing and dispatching efficiently and predictably. Section 4.3 outlines a de layered demultiplexing mechanism and a real time upcall mechanism that processes CORBA requests predictably regardless of the number of active connections, application level target ....

[Article contains additional citation context not shown here]

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing, " in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

....algorithm will thereby first try to distribute flows to ANPEs connected to the same switch port before diverting to other ANPEs on the switch, thus saving switch bandwidth. We have already developed operating system support for ensuring QoS guarantees in the context of two other projects [18, 32] funded in part by DARPA s Quorum. We plan to use these operating system mechanisms and possibly expand them to implement the RC. Other components shown in Figure 5 are already implemented in the Router Plugin platform and briefly described here: the IPv4 IPv6 stack is a dual stack IP ....

Gopalakrishnan, R., Parulkar, G., "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing",In Proceedings of the 1996 ACM SIGMETRICS", May 1996

....such as CPUs, memory, storage throughput, network adapter throughput, and network connection bandwidth and latency. For instance, OS scheduling mechanisms must allow high priority CORBA requests to run to completion and prevent them from being blocked indefinitely by lower priority operations [7, 8]. Section 4.1 describes an OS I O subsystem and network adapter that can provide end to end gigabit data rates and 10 msec latency to CORBA applications. ffl Optimized real time communication protocols: The throughput, latency, and reliability requirements of multimedia applications like ....

....the layered demultiplexing and dispatching mechanisms in conventional ORBs neither schedule nor prioritize demultiplexing behavior. Therefore, the ORB endsystem must provide mechanisms (such as packet filters [9] de layered protocol stacks [10] direct demultiplexing [11] and real time upcalls [7]) that perform CORBA request demultiplexing and dispatching efficiently and predictably. Section 4.3 outlines a de layered demultiplexing mechanism and a real time upcall mechanism that processes CORBA requests predictably regardless of the number of active connections, application level target ....

[Article contains additional citation context not shown here]

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing, " in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

....such as CPUs, memory, storage throughput, network adapter throughput, and network connection bandwidth and latency. For instance, OS scheduling mechanisms must allow high priority CORBA requests to run to completion and prevent them from being blocked indefinitely by lower priority operations [6, 7]. Section 4.1 describes an OS I O subsystem and network adapter that can provide end to end gigabit data rates and 10 msec latency to CORBA applications. ffl Optimized real time communication protocols: The throughput, latency, and reliability requirements of multimedia applications like ....

....the layered demultiplexing and dispatching mechanisms in conventional ORBs neither schedule nor prioritize demultiplexing behavior. Therefore, the ORB endsystem must provide mechanisms (such as packet filters [8] de layered protocol stacks [9] direct demultiplexing [10] and real time upcalls [6]) that perform CORBA request demultiplexing and dispatching efficiently and predictably. Section 4.3 outlines a de layered demultiplexing mechanism and a real time upcall mechanism that processes CORBA requests predictably regardless of the number of active connections, application level target ....

[Article contains additional citation context not shown here]

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing, " in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

....to specify a flexible bandwidth requirement. The scheduler ensures that at least one, and up to B p , pdus will be processed in a period, depending on the system load. 3. 2 The RMDP Scheduling Policy A complete analysis of Rate Monotonic with Delayed Preemption (rmdp) scheduling can be found in [17]. We illustrate this policy only briefly to aid understanding of some of our experimental results. Figure 2 shows the rmdp scheduling algorithm. It is designed to exploit the iterative nature of multimedia processing by treating each iteration as an atomic operation. Preemptions only occur at the ....

R. Gopalakrishnan and G.M. Parulkar. Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing. ACM SIGMETRICS, May 1996.

No context found.

R. Gopalakrishnan and G. M. Parulkar. Bringing real-time scheduling theory and practice closer for multimedia computing. In ACM SIGMETRICS Conference (Philadelphia, PA), pages 1--12, May 1996. 13

No context found.

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing," in SIGMETRICS Conference, (Philadelphia, PA), ACM, May 1996.

No context found.

R. Gopalakrishnan and G. Parulkar, "Bringing Real-time Scheduling Theory and Practice Closer for Multimedia Computing, " in SIGMETRICS Conference, ACM, (Philadelphia, PA), May 1996.

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