Calculating an upper bound on the finishing time of a group of threads executing on a GPU: a preliminary case study

Graphics processor units (GPUs) today can be used for computations that go beyond graphics and such use can attain a performance that is orders of magnitude greater than a normal processor. The software executing on a graphics processor is composed of a set of (often thousands of) threads which operate on different parts of the data and thereby jointly compute a result which is delivered to another thread executing on the main processor. Hence the response time of a thread executing on the main processor is dependent on the finishing time of the execution of threads executing on the GPU. Therefore, we present a simple method for calculating an upper bound on the finishing time of threads executing on a GPU, in particular NVIDIA Fermi. Developing such a method is nontrivial because threads executing on a GPU share hardware resources at very fine granularity.

Analyzing TDMA with slot skipping

Distributed real-time systems, such as factory automation systems, require that computer nodes communicate with a known and low bound on the communication delay. This can be achieved with traditional time division multiple access (TDMA). But improved flexibility and simpler upgrades are possible through the use of TDMA with slot-skipping (TDMA/SS), meaning that a slot is skipped whenever it is not used and consequently the slot after the skipped slot starts earlier. We propose a schedulability analysis for TDMA/SS. We assume knowledge of all message streams in the system, and that each node schedules messages in its output queue according to deadline monotonic. Firstly, we present a non-exact (but fast) analysis and then, at the cost of computation time, we also present an algorithm that computes exact queuing times.

A scalable and efficient approach for obtaining measurements in CAN-Based control systems

The availability of small inexpensive sensor elements enables the employment of large wired or wireless sensor networks for feeding control systems. Unfortunately, the need to transmit a large number of sensor measurements over a network negatively affects the timing parameters of the control loop. This paper presents a solution to this problem by representing sensor measurements with an approximate representation-an interpolation of sensor measurements as a function of space coordinates. A priority-based medium access control (MAC) protocol is used to select the sensor messages with high information content. Thus, the information from a large number of sensor measurements is conveyed within a few messages. This approach greatly reduces the time for obtaining a snapshot of the environment state and therefore supports the real-time requirements of feedback control loops.

New schedulability analysis for WiDom

WiDom is a wireless prioritized medium access control (MAC) protocol which offers a very large number of priority levels. Hence, it brings the potential for employing non-preemptive static-priority scheduling and schedulability analysis for a wireless channel assuming that the overhead of WiDom is modeled properly. One schedulability analysis for WiDom has already been proposed but recent research has created a new version of WiDom with lower overhead (we call it: WiDom with a master node) and for this version of WiDom no schedulability analysis exists. Also, common to the previously proposed schedulability analyses for WiDom is that they cannot analyze message streams with release jitter. Therefore, in this paper we propose a new schedulability analysis for WiDom (with a master node). We also extend the WiDom analyses (with and without master node) to work also for message streams with release jitter.

Elements of scalable data processing

Cooperating objects (COs) is a recently coined term used to signify the convergence of classical embedded computer systems, wireless sensor networks and robotics and control. We present essential elements of a reference architecture for scalable data processing for the CO paradigm.

WiDom: a dominance protocol for wireless medium access

In this paper, we address the problem of sharing a wireless channel among a set of sporadic message streams where a message stream issues transmission requests with real-time deadlines. We propose a collision-free wireless medium access control (MAC) protocol which implements static-priority scheduling, supports a large number of priority levels and is fully distributed. It is an adaptation to a wireless channel of the dominance protocol used in the CAN bus. But, unlike that protocol, our protocol does not require a node having the ability to receive an incoming bit from the channel while transmitting to the channel. The evaluation of the protocol with real embedded computing platforms is presented to show that the proposed protocol is in fact collision-free and prioritized. We measure the response times of our implementation and show that the response-time analysis developed for the protocol offers an upper bound on the response times.

Exploiting a prioritized MAC protocol to efficiently compute min and max in multihop networks

Consider a wireless sensor network (WSN) where a broadcast from a sensor node does not reach all sensor nodes in the network; such networks are often called multihop networks. Sensor nodes take sensor readings but individual sensor readings are not very important. It is important however to compute aggregated quantities of these sensor readings. The minimum and maximum of all sensor readings at an instant are often interesting because they indicate abnormal behavior, for example if the maximum temperature is very high then it may be that a fire has broken out. We propose an algorithm for computing the min or max of sensor reading in a multihop network. This algorithm has the particularly interesting property of having a time complexity that does not depend on the number of sensor nodes; only the network diameter and the range of the value domain of sensor readings matter.

Competitive analysis of partitioned scheduling on uniform multiprocessors

Consider the problem of scheduling a set of sporadically arriving tasks on a uniform multiprocessor with the goal of meeting deadlines. A processor p has the speed Sp. Tasks can be preempted but they cannot migrate between processors. We propose an algorithm which can schedule all task sets that any other possible algorithm can schedule assuming that our algorithm is given processors that are three times faster.

Static-priority scheduling over wireless networks with multiple broadcast domains

We propose a wireless medium access control (MAC) protocol that provides static-priority scheduling of messages in a guaranteed collision-free manner. Our protocol supports multiple broadcast domains, resolves the wireless hidden terminal problem and allows for parallel transmissions across a mesh network. Arbitration of messages is achieved without the notion of a master coordinating node, global clock synchronization or out-of-band signaling. The protocol relies on bit-dominance similar to what is used in the CAN bus except that in order to operate on a wireless physical layer, nodes are not required to receive incoming bits while transmitting. The use of bit-dominance efficiently allows for a much larger number of priorities than would be possible using existing wireless solutions. A MAC protocol with these properties enables schedulability analysis of sporadic message streams in wireless multihop networks.

Competitive analysis of static-priority partitioned scheduling on uniform multiprocessors

Consider the problem of scheduling a set of sporadically arriving tasks on a uniform multiprocessor with the goal of meeting deadlines. A processor p has the speed Sp. Tasks can be preempted but they cannot migrate between processors. On each processor, tasks are scheduled according to rate-monotonic. We propose an algorithm that can schedule all task sets that any other possible algorithm can schedule assuming that our algorithm is given processors that are √2 / √2−1 ≈ 3.41 times faster. No such guarantees are previously known for partitioned static-priority scheduling on uniform multiprocessors.

Exploiting a prioritized MAC protocol to efficiently compute interpolations

Consider a network where all nodes share a single broadcast domain such as a wired broadcast network. Nodes take sensor readings but individual sensor readings are not the most important pieces of data in the system. Instead, we are interested in aggregated quantities of the sensor readings such as minimum and maximum values, the number of nodes and the median among a set of sensor readings on different nodes. In this paper we show that a prioritized medium access control (MAC) protocol may advantageously be exploited to efficiently compute aggregated quantities of sensor readings. In this context, we propose a distributed algorithm that has a very low time and message-complexity for computing certain aggregated quantities. Importantly, we show that if every sensor node knows its geographical location, then sensor data can be interpolated with our novel distributed algorithm, and the message-complexity of the algorithm is independent of the number of nodes. Such an interpolation of sensor data can be used to compute any desired function; for example the temperature gradient in a room (e.g., industrial plant) densely populated with sensor nodes, or the gas concentration gradient within a pipeline or traffic tunnel.

A two-competitive approximate schedulability analysis of CAN

Consider the problem of deciding whether a set of n sporadic message streams meet deadlines on a Controller Area Network (CAN) bus for a specified priority assignment. It is assumed that message streams have implicit deadlines and no release jitter. An algorithm to solve this problem is well known but unfortunately it time complexity is non-polynomial. We present an algorithm with polynomial time-complexity for computing an upper bound on the response times. Clearly, if the upper bound on the response time does not exceed the deadline then all deadlines are met. The pessimism of our approach is proven: if the upper bound of the response time exceeds the deadline then the response time exceeds the deadline as well for a CAN network with half the speed.

Exact analysis of TDMA with slot skipping

Consider a communication medium shared among a set of computer nodes; these computer nodes issue messages that are requested to be transmitted and they must finish their transmission before their respective deadlines. TDMA/SS is a protocol that solves this problem; it is a specific type of Time Division Multiple Access (TDMA) where a computer node is allowed to skip its time slot and then this time slot can be used by another computer node. We present an algorithm that computes exact queuing times for TDMA/SS in conjunction with Rate-Monotonic (RM) or Earliest- Deadline-First (EDF).

Delay-bounded medium access for unidirectional wireless links

Consider a wireless network where links may be unidirectional, that is, a computer node A can broadcast a message and computer node B will receive this message but if B broadcasts then A will not receive it. Assume that messages have deadlines. We propose a medium access control (MAC) protocol which replicates a message in time with carefully selected pauses between replicas, and in this way it guarantees that for every message at least one replica of that message is transmitted without collision. The protocol ensures this with no knowledge of the network topology and it requires neither synchronized clocks nor carrier sensing capabilities. We believe this result is significant because it is the only MAC protocol that offers an upper bound on the message queuing delay for unidirectional links without relying on synchronized clocks.

Linearizability and schedulability

Consider the problem of scheduling a set of tasks on a single processor such that deadlines are met. Assume that tasks may share data and that linearizability, the most common correctness condition for data sharing, must be satisfied. We find that linearizability can severely penalize schedulability. We identify, however, two special cases where linearizability causes no or not too large penalty on schedulability.