Outstanding Requests Research Articles

Equivalence of two linear programming relaxations for broadcast scheduling*1

A server needs to compute a broadcast schedule for n pages whose request times are known in advance. Outputting a page satisfies all outstanding requests for the page. The goal is to minimize the average waiting time of a client. In this paper, we show the equivalence of two apparently different relaxations that have been considered for this problem.

Interposed proportional sharing for a storage service utility

This paper develops and evaluates new share-based scheduling algorithms for differentiated service quality in network services, such as network storage servers. This form of resource control makes it possible to share a server among multiple request flows with probabilistic assurance that each flow receives a specified minimum share of a server's capacity to serve requests. This assurance is important for safe outsourcing of services to shared utilities such as Storage Service Providers.Our approach interposes share-based request dispatching on the network path between the server and its clients. Two new scheduling algorithms are designed to run within an intermediary (e.g., a network switch), where they enforce fair sharing by throttling request flows and reordering requests; these algorithms are adaptations of Start-time Fair Queuing (SFQ) for servers with a configurable degree of internal concurrency. A third algorithm, Request Windows (RW), bounds the outstanding requests for each flow independently; it is amenable to a decentralized implementation, but may restrict concurrency under light load. The analysis and experimental results show that these new algorithms can enforce shares effectively when the shares are not saturated, and that they provide acceptable performance isolation under saturation. Although the evaluation uses a storage service as an example, interposed request scheduling is non-intrusive and views the server as a black box, so it is useful for complex services with no internal support for differentiated service quality.

Equivalence of two linear programming relaxations for broadcast scheduling

A server needs to compute a broadcast schedule for n pages whose request times are known in advance. Outputting a page satisfies all outstanding requests for the page. The goal is to minimize the average waiting time of a client. In this paper, we show the equivalence of two apparently different relaxations that have been considered for this problem.

Algorithms for Minimizing Response Time in Broadcast Scheduling

In this paper we study the following problem. There are n pages which clients can request at any time. The arrival times of requests for pages are known in advance. Several requests for the same page may arrive at different times. There is a server that needs to compute a good broadcast schedule. Outputting a page satisfies all outstanding requests for the page. The goal is to minimize the average waiting time of a client. This problem has recently been shown to be NP-hard. For any fixed α, 0 < α \le ½, we give a 1/α-speed, polynomial time algorithm with an approximation ratio of 1/(1 – α). For example, setting α = ½ gives a 2-speed, 2-approximation algorithm. In addition, we give a 4-speed, 1-approximation algorithm improving the previous bound of 6-speed, 1-approximation algorithm.

Multicast Pull Scheduling: When Fairness Is Fine

We investigate server scheduling policies to minimize average user perceived latency in pull-based client-server systems (systems where multiple clients request data from a server) where the server answers requests on a multicast/ broadcast channel. We first show that there is no O(1) -competitive algorithm for this problem. We then give a method to convert any nonclairvoyant unicast scheduling algorithm A to nonclairvoyant multicast scheduling algorithm B . We show that if A works well, when jobs can have parallel and sequential phases, then B works well if it is given twice the resources. More formally, if A is an s -speed c -approximation unicast algorithm, then its counterpart algorithm B is a 2s -speed c -approximation multicast algorithm. It is already known [5] that Equi-partition, which devotes an equal amount of processing power to each job, is a (2 + e) -speed O(1 + 1/e) -approximation algorithm for unicast scheduling of such jobs. Hence, it follows that the algorithm {BEQUI}, which broadcasts all requested files at a rate proportional to the number of outstanding requests for that file, is a (4 + e) -speed O(1 + 1/e) -approximation algorithm. We give another algorithm BEQUI-EDF and show that BEQUI-EDF is also a (4 + e) -speed O(1 + 1/e) -approximation algorithm. However, BEQUI-EDF has the advantage that the maximum number of preemptions is linear in the number of requests, and the advantage that no preemptions occur if the data items have unit size.

Anticipatory scheduling

Disk schedulers in current operating systems are generally work-conserving, i.e., they schedule a request as soon as the previous request has finished. Such schedulers often require multiple outstanding requests from each process to meet system-level goals of performance and quality of service. Unfortunately, many common applications issue disk read requests in a synchronous manner, interspersing successive requests with short periods of computation. The scheduler chooses the next request too early; this induces deceptive idleness, a condition where the scheduler incorrectly assumes that the last request issuing process has no further requests, and becomes forced to switch to a request from another process.We propose the anticipatory disk scheduling framework to solve this problem in a simple, general and transparent way, based on the non-work-conserving scheduling discipline. Our FreeBSD implementation is observed to yield large benefits on a range of microbenchmarks and real workloads. The Apache webserver delivers between 29% and 71% more throughput on a disk-intensive workload. The Andrew filesystem benchmark runs faster by 8%, due to a speedup of 54% in its read-intensive phase. Variants of the TPC-B database benchmark exhibit improvements between 2% and 60%. Proportional-share schedulers are seen to achieve their contracts accurately and efficiently.

Evaluating the impact of locality on the performance of large-scale SCI multiprocessors

Hierarchical ring-based multiprocessor systems are attractive and enjoy several advantages over other type of systems. They ensure unique paths between nodes, simple node interfaces and simple cross-ring connections. Furthermore, employing point-to-point links allows the system to run at high clock rate which increases bandwidth and decreases latency. This paper investigates the performance of hierarchical ring-based shared-memory multiprocessors. Rings in the hierarchy are composed of point-to-point, unidirectional links and apply the Scalable Coherent Interface (SCI) protocol. We pay special emphasis on the impact of locality on processor and interconnection design issues such as number of outstanding requests, and ring topology. We find that in order to exploit the power of hierarchical multiprocessors an accurate and appropriate model of locality must be used. Hierarchical multiprocessors that are well balanced (uniform) tend to provide lower latency and higher system throughput. For non-uniform systems, high degree of locality is required for the hierarchies to perform well. However, restricting the number of outstanding transactions per processor is important in decreasing packets latency and avoiding network contention.

Effect of non-uniform memory request on the performance of buffered multiprocessor systems

Performance analysis of multiple-bus systems is usually carried out under the assumption of a uniform memory request model. Hot spots arising in multiprocessor systems give rise to non-uniform memory requests. It is known that a hot spot memory request pattern results in a significant degradation in the performance of a buffered multistage interconnection network. The aim of this research is to study the effect of hot spots on the performance of buffered multiple-bus systems, and to compare the performance of buffered and unbuffered systems. Analytical models based on Markov chains have been developed to determine the bandwidth of buffered multiple-bus systems in the presence of a hot spot memory request pattern. The models assume that unsuccessful memory requests are queued in the buffers at the memory modules. Furthermore, processors with outstanding requests are not allowed to generate new requests. The model allows a fast and inexpensive method to evaluate the performance of a buffered multiple-bus system in the presence of a hot spot memory request pattern.

Modeling and Simulation of a Hierarchical, Distributed, Dynamic Inventory Management Scheme

This paper proposes a Hierarchical, Distributed, Dynamic, Inventory management (HDDI) scheme. HDDI introduces the concept of emergency replenishment, wherein a retail unit is permitted to request the transfer of items from another retail unit in the neighborhood when, despite an outstanding reorder request from the warehouse, the demand continues to rapidly deplete the inventory level and the latter falls below an emergency threshold value. To be effective, with today's rapid price changes and fast-paced consumer demand, the cost function in HDDI is dynamic, i.e., reevaluated with current system parameters whenever the inventory level falls below the threshold. For fast results, the overall inventory management computation is distrubuted among all of the retail units. HDDI is modeled for a few representative inventory management networks, simulated on 17+ SUN workstations for stochastically generated demand functions and for different sets of values of the key parameters, and the performance results are reported.

A data request scheduling policy for PCS networks

This paper proposes a scheduling policy called latest preempted, earliest resumed (LPER) for personal communications services (PCS) systems that support voice services as well as the circuit mode data services. The policy gives priority to the voice services over the data services. The scheduling of voice requests is performed as if no data requests exist. Thus, if no idle channel is available when a voice request arrives, a data channel (if exists) is interrupted and the channel will be used for the voice request. The interrupted data request will be resumed when an idle channel is available. In LPER, the data channel selected for interruption is the one that serves the latest outstanding data request (i.e., other outstanding requests enter the system earlier than this request). When an occupied channel is released, LPER resumes the earliest request that entered the system. An analytic model is proposed to study the performance of LPER and provides guidelines to select the input parameters for the PCS systems.

Dynamic batching policies for an on-demand video server

In a video-on-demand environment, continuous delivery of video streams to the clients is guaranteed by sufficient reserved network and server resources. This leads to a hard limit on the number of streams that a video server can deliver. Multiple client requests for the same video can be served with a single disk I/O stream by sending (multicasting) the same data blocks to multiple clients (with the multicast facility, if present in the system). This is achieved by batching (grouping) requests for the same video that arrive within a short time. We explore the role of customerwaiting time and reneging behavior in selecting the video to be multicast. We show that a first come, first served (FCFS) policy that schedules the video with the longest outstanding request can perform better than the maximum queue length (MQL) policy that chooses the video with the maximum number of outstanding requests. Additionally, multicasting is better exploited by scheduling playback of the n most popular videos at predetermined, regular intervals (hence, termed FCFS-n). If user reneging can be reduced by guaranteeing that a maximum waiting time will not be exceeded, then performance of FCFS-n is further improved by selecting the regular playback intervals as this maximum waiting time. For an empirical workload, we demonstrate a substantial reduction (of the order of 60%) in the required server capacity by batching.

A Scalable Shared Queue on a Distributed Memory Machine

The emergence of low latency, high throughput routers means that network locality issues no longer dominate the performance of parallel algorithms. One of the key performance issues is now the even distribution of work across the machine, as the problem size and number of processors increase. This paper describes the implementation of a highly scalable shared queue, supporting the concurrent insertion and deletion of elements. The main characteristics of the queue are that there is no fixed limit on the number of outstanding requests and the performance scales linearly with the number of processors (subject to increasing network latencies). The queue is implemented using a general-purpose computational model, called the WPRAM. The model includes a shared address space which uses weak coherency semantics. The implementation makes extensive use of pairwise synchronization and concurrent atomic operations to achieve scalable performance. The WPRAM is targeted at the class of distributed memory machines which use a scalable interconnection network.

Performance analysis of finite-buffered asynchronous multistage interconnection networks

We present a queueing model for performance analysis of finite-buffered multistage interconnection networks. The proposed model captures network behaviour in an asynchronous communication mode and is based on realistic assumptions. A uniform traffic model is developed first and then extended to capture nonuniform traffic in the presence of a hot-spot. Throughput and delay are computed using the proposed model and the results are validated via simulation. The analysis is extended to predict performance of MIN-based multiprocessors. The effects of buffer length, switch size, and the maximum allowable outstanding requests on the system performance are discussed. Various design decisions using this model are drawn with respect to delay, throughput, and system power.

Dynamic self-invalidation

This paper introduces dynamic self-invalidation (DSI), a new technique for reducing cache coherence overhead in shared-memory multiprocessors. DSI eliminates invalidation messages by having a processor automatically invalidate its local copy of a cache block before a conflicting access by another processor. Eliminating invalidation overhead is particularly important under sequential consistency, where the latency of invalidating outstanding copies can increase a program's critical path.DSI is applicable to software, hardware, and hybrid coherence schemes. In this paper we evaluate DSI in the context of hardware directory-based write-invalidate coherence protocols. Our results show that DSI reduces execution time of a sequentially consistent full-map coherence protocol by as much as 41%. This is comparable to an implementation of weak consistency that uses a coalescing write-buffer to allow up to 16 outstanding requests for exclusive blocks. When used in conjunction with weak consistency, DSI can exploit tear-off blocks---which eliminate both invalidation and acknowledgment messages---for a total reduction in messages of up to 26%.

MD3Q: A distributed queueing protocol with full channel capacity re-use and guarantee of bandwidth

Abstract This paper describes an enhanced distributed queueing protocol called Dynamic Distributed Dual Queue (D3Q). This protocol incorporates destination release and request cancellation. Therefore D3Q allows reuse of channel capacity and is suitable for both bus and ring architectures. After that, Multirequest D3Q (MD3Q) is introduced, which is an enhanced version of D3Q incorporating multiple outstanding requests. This protocol version includes new features for guaranteeing bandwidth to a node and for supporting multiple priorities. MD3Q is not merely an enhancement of D3Q or DQDB but rather a generalisation for Distributing Queueing (DQ) protocols. The described methods of bandwidth re-use and guarantee of bandwidth can be generally applied to all kinds of distributed queueing protocols. By using only 4 bits for the access protocol and supporting multiple levels of priority, this protocol is very suitable for performing Generic Flow Control (GFC) at the ATM User-to-Network Interface (UNI), where it is currently being studied further in CCITT SG XVIII.

Simulations with SCI as a data carrier in data acquisition systems

Detailed simulations of processor networks based on the Scalable Coherent Interface (SCI) show that SCI is suitable as a data carrier in data acquisition systems where the total bandwidth need is in the multi GBytes/s range and a low latency is required. The objective of these simulations was to find topologies with low latency and high bandwidth, but also with the cost of implementation in mind. A ring-to-ring bridge has been used as the building element for the networks. The simulations have been performed on regular k-ary n-cube type topologies from a few tens of nodes and up to about 500 nodes under different load conditions. Among the parameters which have been manipulated in the simulations are the number of nodes, topology structure, number of outstanding requests and load in the system.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Extended distributed queueing for integrated services

A method for supporting time-constrained and nontime-constrained integrated services on a shared dual bus network is presented. The method is an extension of the basic distributed queue protocol defined in the IEEE 802.6 distributed-queue dual-bus (DQDB) MAN standard. The extended distributed queueing protocol differs from the basic protocol in allowing more than one outstanding request in the distributed queue at any one time. Used at the high priority and in conjunction with segment stream shaping and subscribed service limiting, the extended protocol provides bearer services with guaranteed bandwidth and bounded delays. Bearer services with fair but nonguaranteed bandwidth shares, suitable for nontime-constrained applications, are provided by the same protocol and bandwidth balancing operating at the lowest priority. A number of the desirable attributes of the proposed scheme are identified including the optimality of the delay variance introduced by the access protocol. An approximate analysis of the access delay experience by time-constrained streams is presented. >

DISTRIBUTED DEADLOCK DETECTION ALGORITHMS

A distributed system consists of a set of loosely connected state machines which do not share a global memory. The global state of the system depends on the state of each process in the system. The set of global states can be split up into two categories, legal and illegal. This paper deals with methods of detecting deadlocks in distributed systems. One way that has been used to detect deadlocks is by sending probes around the system. If a process thinks that it may be deadlocked, it initiates a probe. If the probe is received by the initiator, the initiator declares deadlock. This paper uses the idea of states of processes In order to detect the deadlock. The algorithm runs continually and does not have to be initiated. This paper presents deadlock detection algorithms for single and multiple outstanding requests. A method for deadlock resolution is also discussed. The algorithms detect all deadlocks and do not detect false deadlocks.

Integration of knowledge and method in real-world discovery

An architecture which controls automated discovery of empirical knowledge must integrate a robotic component which interacts with the real world, knowledge network which accumulates knowledge, a discovery method, and the currently executed discovery tasks. We describe such an architecture implemented in the FAHRENHEIT system, concentrating on the integration and interaction between elements. The robotic component includes hardware connections to the external instruments and manipulators, operational procedures to control the hardware, and a process that controls outstanding requests for measurements and manipulations. Knowledge network describes the actual state of knowledge about the world, including all discoveries that have been made. The state of knowledge expands as the discovery process continues. Operational procedures are expanded and refined as a result of discoveries. Better procedures, in turn, allow FAHRENHEIT to collect better data and to improve its knowledge. The discovery method consists of a network of generic goals and plans for goal execution. Discovery goals require search, so most of the generic plans are solution schemes to search problems. All search problems are defined by the application of a uniform search template. Actual goals and plans executed at any given time form a recursive network of instances of generic goals and plans. The instances are selected based on the generic network of goals and plans, and the current state of knowledge. The selected plans are search instances. At any given time a number of search instances of various types may be in progress, each of them plan leading to new discoveries.

Performance evaluation of MiniMAP networks

A simulation study of the performances of MiniMAP networks is presented. The two network models used in the simulation take into account the effect of the maximum number of outstanding requests set for each manufacturing message specification (MMS) application association. Three different situations are considered. In the first, all the messages are generated at the same priority level, whereas in the other two, four priority classes that are directly mapped onto channel priorities are introduced. However, the second situation considers the use of priorities only on the client side of MMS, whereas the third uses priorities on both sides.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>