Load Balancing Performance in Distributed Storage with Regular Balanced Redundancy

Abstract

Contention at the storage nodes is the main cause of long and variable data access times in distributed storage systems. Offered load on the system must be balanced across the storage nodes in order to minimize contention, and load balance in the system should be robust against the skews and fluctuations in content popularities. Data objects are replicated across multiple nodes in practice to allow for load balancing. However redundancy increases the storage requirement and should be used efficiently. We evaluate load balancing performance of natural storage schemes in which each data object is stored at d different nodes and each node stores the same number of objects. We find that load balance in a system of n nodes improves multiplicatively with d as long as d = o (log(n)), and improves exponentially as soon as d = Θ (log(n)). We show that the load balance in the system improves the same way with d when the service choices are created with XOR's of r objects rather than object replicas, which also reduces the storage overhead multiplicatively by r. However, unlike accessing an object replica, access through a recovery set composed by an XOR'ed object copy requires downloading content from r nodes, which increases the load imbalance in the system additively by r.

Service Rate Region: A New Aspect of Coded Distributed System Design

“…Our current throughput-optimal allocation policies such as water-filling are biased towards the later strategy because they give higher priority to sending requests to smaller recovery groups. An alternative approach is to design policies that perform the best possible load balancing of requests across the nodes, that is, minimize the cumulative request rate assigned to the maximally loaded node, as recently considered in [79]. Such load-balancing strategies can give better latency performance than the throughput-optimal strategies considered in this paper.…”

A Combinatorial View of the Service Rates of Codes Problem, its Equivalence to Fractional Matching and its Connection with Batch Codes

“…We note now and explain in detail later that because of the constraints on the service rate of servers, by maximizing the service capacity the load balancing is provided in the distributed storage system (see [6]).…”

“…Proof of Corollary 1: An instance of the maximum fractional matching vector in the graph representation of an [n, k] q code can be obtained by solving the following LP according to (6).…”

A Geometric View of the Service Rates of Codes Problem and its Application to the Service Rate of the First Order Reed-Muller Codes

“…The most related to this work are papers concerned with content download from coded storage. Load balancing in such systems has recently been addressed in [13]. Memory allocation that maximizes the probability of successful content download under limited access to distributed storage was considered in e.g., [14], [15] and references therein.…”