Paolo Valente

High Throughput Disk Scheduling with Fair Bandwidth Distribution

Mainstream applications-such as file copy/transfer, Web, DBMS, or video streaming-typically issue synchronous disk requests. As shown in this paper, this fact may cause work-conserving schedulers to fail both to enforce guarantees and provide a high disk throughput. A high throughput can be, however, recovered by just idling the disk for a short time interval after the completion of each request. In contrast, guarantees may still be violated by existing time-stamp-based schedulers because of the rules they use to tag requests. Budget Fair Queuing (BFQ), the new disk scheduler presented in this paper, is an example of how disk idling, combined with proper back-shifting of request time stamps, may allow a time-stamp-based disk scheduler to preserve both guarantees and a high throughput. Under BFQ, each application is always guaranteed-over any time interval and independently of whether it issues synchronous requests-a bounded lag with respect to its reserved fraction of the total number of bytes transferred by the disk device. We show the single-disk performance of our implementation of BFQ in the Linux kernel through experiments with real and emulated mainstream applications.

An upper bound to the lateness of soft real-time tasks scheduled by EDF on multiprocessors

Multiprocessors are now commonplace for efficiently achieving high computational power, even in embedded systems. A considerable research effort is being addressed to schedulability analysis of global scheduling in symmetric multiprocessor platforms (SMP), where there is a global queue of ready tasks, and preemption and migration are allowed. In many soft real-time applications (as e.g. multimedia and telecommunication) a bounded lateness is often tolerated. Unfortunately, when considering priority-driven scheduling of periodic/sporadic tasks, previous results only focused on guaranteeing all deadlines, and provided worst-case utilization bounds that are lower than the maximum available computational power. In particular, until now, the existence of an upper bound on the lateness of soft real-time tasks for a fully utilized SMP was still an open problem. In this paper we do solve this problem by providing an upper bound to the lateness of periodic/sporadic tasks - with relative deadlines equal to periods/minimum inter-arrival times - scheduled by EDF on a SMP, under the only assumption that the total utilization is no higher than the total system capacity

Exact GPS simulation with logarithmic complexity, and its application to an optimally fair scheduler

Generalized Processor Sharing (GPS) is a fluid scheduling policy providing perfect fairness. The minimum deviation (lead/lag) with respect to the GPS service achievable by a packet scheduler is one packet size. To the best of our knowledge, the only packet scheduler guaranteeing such minimum deviation is Worst-case Fair Weighted Fair Queueing (WF2Q), that requires on-line GPS simulation. Existing algorithms to perform GPS simulation have O(N) complexity per packet transmission (N being the number of competing flows). Hence WF2Q has been charged for O(N) complexity too. Schedulers with lower complexity have been devised, but at the price of at least O(N) deviation from the GPS service, which has been shown to be detrimental for real-time adaptive applications and feedback based applications. Furthermore, it has been proven that the lower bound complexity to guarantee O(1) deviation is Ω(log N), yet a scheduler achieving such result has remained elusive so far.In this paper we present an algorithm that performs exact GPS simulation with O(log N) worst-case complexity and small constants. As such it improves the complexity of all the packet schedulers based on GPS simulation. In particular, using our algorithm within WF2Q, we achieve the minimum deviation from the GPS service with O(log N) complexity, thus matching the aforementioned lower bound. Furthermore, we assess the effectiveness of the proposed solution by simulating real-world scenarios.

Exact GPS simulation and optimal fair scheduling with logarithmic complexity

Generalized processor sharing (GPS) is a fluid scheduling policy providing perfect fairness over both constant-rate and variable-rate links. The minimum deviation (lead/lag) with respect to the GPS service achievable by a packet scheduler is one maximum packet size. To the best of our knowledge, the only packet scheduler guaranteeing the minimum deviation is worst-case fair weighted fair queueing , which requires on-line GPS simulation. Existing algorithms to perform GPS simulation have worst-case computational complexity per packet transmission (being the number of competing flows). Hence, has been charged for complexity too. However it has been proven that the lower bound complexity to guarantee deviation is, yet a scheduler achieving such a result has remained elusive so far. In this paper, we present L-GPS, an algorithm that performs exact GPS simulation with worst-case complexity and small constants. As such it improves the complexity of all the packet schedulers based on GPS simulation. We also present , an implementation of based on L-GPS. has complexity with small constants, and, since it achieves the minimum possible deviation, it does match the aforementioned complexity lower bound. Furthermore, both L-GPS and comply with constant-rate as well as variable-rate links. We assess the effectiveness of both algorithms by simulating real-world scenarios.

QFQ: efficient packet scheduling with tight guarantees

Packet scheduling, together with classification, is one of the most expensive processing steps in systems providing tight bandwidth and delay guarantees at high packet rates. Schedulers with near-optimal service guarantees and O(1) time complexity have been proposed in the past, using techniques such as timestamp rounding and flow grouping to keep their execution time small. However, even the two best proposals in this family have a per-packet cost component that is linear either in the number of groups or in the length of the packet being transmitted. Furthermore, no studies are available on the actual execution time of these algorithms. In this paper we make two contributions. First, we present Quick Fair Queueing (QFQ), a new O( 1) scheduler that provides near-optimal guarantees and is the first to achieve that goal with a truly constant cost also with respect to the number of groups and the packet length. The QFQ algorithm has no loops and uses very simple instructions and data structures that contribute to its speed of operation. Second, we have developed production-quality implementations of QFQ and of its closest competitors, which we use to present a detailed comparative performance analysis of the various algorithms. Experiments show that QFQ fulfills our expectations, outperforming the other algorithms in the same class. In absolute terms, even on a low-end workstation, QFQ takes about 110 ns for an enqueue()/dequeue() pair (only twice the time of DRR, but with much better service guarantees).

A low-latency and high-throughput scheduler for emergency and wireless networks

Providing QoS guarantees, boosting throughput and saving energy over wireless links is a challenging task, especially in emergency networks, where all of these features are crucial during a disaster event. A common solution is using a single, integrated scheduler that deals both with the QoS guarantees and the wireless link issues. Unfortunately, such an approach is not flexible and does not allow any of the existing high-quality schedulers for wired links to be used without modifications. We address these issues through a modular architecture which permits the use of existing packet schedulers for wired links over wireless links, as they are, and at the same time allows the flexibility to adapt to different channel conditions. We validate the effectiveness of our modular architecture by showing, through formal analysis as well as experimental results, that this architecture enables us to get a new scheduler with the following features, by just combining existing schedulers: execution time and energy consumption close to that of just a Deficit Round Robin, accurate fairness and low latency, possibility to set the desired trade-off between throughput-boosting level and granularity of service guarantees, by changing one parameter. In particular, we show that this scheduler, which we named Highth-roughput Twin Fair scheduler (HFS), outperforms one of the most accurate and efficient integrated schedulers available in the literature.

An STDMA-based framework for QoS provisioning in wireless mesh networks

Providing strong QoS guarantees for wireless multi-hop networks is very challenging, due to many factors such as use of a shared communication medium, variability in wireless link quality, and so on. However, wireless mesh technology gives the opportunity to alleviate some of these problems, due to lack of mobility in the wireless infrastructure, and presence of natural centralization points in the network. The main contribution of this paper is the definition of a simple framework that exploits these features to provide provable, strong QoS guarantees to network clients. In particular, admitted clients are guaranteed a certain minimum bandwidth and maximum delay on their connections. The framework is based on STDMA scheduling at the MAC layer, which is periodically executed at the network manager to adapt to changes in traffic demand. While scheduling computation is centralized, admission control is performed locally at the wireless backbone nodes, thus reducing signaling. We propose two bandwidth distribution and related admission control policies, which are at opposite ends of the network utilization/spatial fairness trade-off. Through extensive simulations, we show that the proposed framework achieves its design goals of providing strong QoS guarantees to VoIP clients while not sacrificing throughput in a realistic mesh network scenario, also in presence of highly unbalanced load at the backbone nodes. To the best of our knowledge, this is the first proposal with similar features for wireless mesh networks.

A memory-centric approach to enable timing-predictability within embedded many-core accelerators

There is an increasing interest among real-time systems architects for multi- and many-core accelerated platforms. The main obstacle towards the adoption of such devices within industrial settings is related to the difficulties in tightly estimating the multiple interferences that may arise among the parallel components of the system. This in particular concerns concurrent accesses to shared memory and communication resources. Existing worst-case execution time analyses are extremely pessimistic, especially when adopted for systems composed of hundreds-tothousands of cores. This significantly limits the potential for the adoption of these platforms in real-time systems. In this paper, we study how the predictable execution model (PREM), a memory-aware approach to enable timing-predictability in realtime systems, can be successfully adopted on multi- and manycore heterogeneous platforms. Using a state-of-the-art multi-core platform as a testbed, we validate that it is possible to obtain an order-of-magnitude improvement in the WCET bounds of parallel applications, if data movements are adequately orchestrated in accordance with PREM. We identify which system parameters mostly affect the tremendous performance opportunities offered by this approach, both on average and in the worst case, moving the first step towards predictable many-core systems.

Reducing the Execution Time of Fair-Queueing Packet Schedulers

Deficit Round Robin (DRR) is probably the most scalable fair-queueing packet scheduler. Unfortunately, it suffers from high delay and jitter with respect to a perfectly fair (and smooth) service. Schedulers providing much better service guarantees exist, but have a higher computational cost. In this paper we deal with this issue by proposing a modification scheme for reducing the amortized execution time of the latter, more accurate schedulers. Modified schedulers preserve guarantees close to the original ones, and can also handle seamlessly both leaves and internal nodes in a hierarchical setting. We also present Quick Fair Queueing Plus (QFQ+), a fast fair-queueing scheduler that we defined using this scheme, and that is now in mainline Linux. On one hand, QFQ+ guarantees near-optimal worst-case deviation with respect to a perfectly fair service. On the other hand, with QFQ+, the time and energy needed to process packets are close to those needed with DRR, and may be even lower than with DRR exactly in the scenarios where the better service properties of QFQ+ make a difference.

Improving application responsiveness with the BFQ disk I/O scheduler

BFQ (Budget Fair Queueing) is a production-quality, proportional-share disk scheduler with a relatively large user base. Part of its success is due to a set of simple heuristics that we added to the original algorithm about one year ago. These heuristics are the main focus of this paper. The first heuristic enriches BFQ with one of the most desirable properties for a desktop or handheld system: responsiveness. The remaining heuristics improve the robustness of BFQ across heterogeneous devices, and help BFQ to preserve a high throughput under demanding workloads. To measure the performance of these heuristics we have implemented a suite of micro and macro benchmarks mimicking several real-world tasks, and have run it on three different systems with a single rotational disk. We have also compared our results against Completely Fair Queueing (CFQ), the default Linux disk scheduler.