Lexicographic QoS scheduling for parallel I/O

Abstract

High-end shared storage systems serving multiple independent workloads must assure that concurrently executing clients will receive a fair or agreed-upon share of system I/O resources. In a parallel I/O system an application makes requests for specific disks at different steps of its computation depending on the data layout and its computational state. Different applications contend for disk access making the problem of maintaining fair allocation challenging.We propose a model for differentiated disk bandwidth allocation based on lexicographic minimization, and provide new efficient scheduling algorithms to allocate the I/O bandwidth fairly among contending applications. A major contribution of our model is its ability to handle multiple parallel disks and contention for disks among the concurrent applications. Analysis and simulation-based evaluation shows that our algorithms provide performance isolation, weighted allocation of resources, and are work conserving. The solutions are also applicable to other shared resource environments dealing with non-uniform heterogeneous servers.

vPFS: Bandwidth virtualization of parallel storage systems

“…The majority of the storage resource schedulers in the literature focuses on the allocation of a single storage resource (e.g., a storage server, device, or a cluster of interchangeable storage resources) and addresses the local throughput or latency objectives. LexAS [32] was proposed for fair bandwidth scheduling on a storage system with parallel disks, but I/Os are not striped and the scheduling is done with a centralized controller. DSFQ [7] is a distributed algorithm that can realize total service proportional sharing across all the storage resources that satisfy workload requests.…”

Performance evaluation of a new scheduling algorithm for distributed systems with security heterogeneity

“…For example, a network-based QoS-aware multimedia application can deliver video frames so that the display is jitter-free to some requested levels [2]. Since scheduling plays a crucial role in achieving high performance for applications in distributed systems, extensive research has been conducted on developing QoS-driven scheduling heuristics since the recent decades [3,5,7,10,11,13,21,30,31]. A common conclusion drawn from these studies is that a high and adaptive QoS can be accomplished through scheduling.…”

d -clock

“…These include: Lexicographic Minimization [2] (provides optimal global fairness, but is centralized, batched, and can result in starvation at individual servers), Centralized Fair Queuing [3] (not distributed, starvation prone) and Local Fair Queuing at servers and Distributed Tagging with Fair Queuing at servers [5] (avoid resource-specific starvation but cannot make any global guarantees). All of these schemes either cause resource specific starvation or do unfair bandwidth allocation.…”

“…Define the weighted global allocation vectorW = [A 1 /B 1 , A 2 /B 2 , ··· , A n /B n ], where A i = ∑ j∈S(i) a i j .A feasible allocation must satisfy (i) for any f i and s j ∈ S(i), a i j ≥ L i j (to avoid resource-specific starvation), and (ii) for any s j , ∑ f i ∈F( j) a i j = C i (work-conserving server capacity constraint). The goal of the model is to find the lexicographically minimum[2] vector, W , among all feasible allocations. Intuitively, this tries to makes the components of vector W equal (A i /B i = A j /B j ), so that the A i 's are in proportion to their global bandwidth requirements, subject to meeting all local requirements.…”