Differentiated Latency in Data Center Networks with Erasure Coded Files Through Traffic Engineering

Abstract

Abstract-This paper proposes an algorithm to minimize weighted service latency for different classes of tenants (or service classes) in a data center network where erasure-coded files are stored on distributed disks/racks and access requests are scattered across the network. Due to limited bandwidth available at both top-of-the-rack and aggregation switches and tenants in different service classes need differentiated services, network bandwidth must be apportioned among different intraand inter-rack data flows for different service classes in line with their traffic statistics. We formulate this problem as weighted queuing and employ a class of probabilistic request scheduling policies to derive a closed-form upper-bound of service latency for erasure-coded storage with arbitrary file access patterns and service time distributions. The result enables us to propose a joint weighted latency (over different service classes) optimization over three entangled "control knobs": the bandwidth allocation at top-of-the-rack and aggregation switches for different service classes, dynamic scheduling of file requests, and the placement of encoded file chunks (i.e., data locality). The joint optimization is shown to be a mixed-integer problem. We develop an iterative algorithm which decouples and solves the joint optimization as 3 sub-problems, which are either convex or solvable via bipartite matching in polynomial time. The proposed algorithm is prototyped in an open-source, distributed file system, Tahoe, and evaluated on a cloud testbed with 16 separate physical hosts in an Openstack cluster using 48-port Cisco Catalyst switches. Experiments validate our theoretical latency analysis and show significant latency reduction for diverse file access patterns. The results provide valuable insights on designing low-latency data center networks with erasure coded storage.

Sprout: A Functional Caching Approach to Minimize Service Latency in Erasure-Coded Storage

“…The SSD partitions come from different SSD drives on one node. The number of placement group in cache tier is 128 from Equation (17). Journal settings and number of placement groups in the storage pool are the same with the optimal caching case.…”

“…In these systems, the rapid growth of data traffic such as those generated by online video streaming, Big Data analytics, social networking and E-commerce activities has put a significant burden on the underlying networks of datacenter storage systems. Many researchers have begun to focus on latency analysis in erasure coded storage systems [7][8][9][10][11][12][13][14] and to investigate algorithms for joint latency optimization and resource management [12,[14][15][16][17][18].…”

Dynamic Security Exchange Scheduling Model for Business Workflow Based on Queuing Theory in Cloud Computing

“…Queuing theory is a mathematical research of waiting lines or queues which focuses on identifying and managing the response time of users for services. Queuing theory was created to describe the Copenhagen telephone exchange originally and the ideas had seen applications including telecommunication [28], traffic engineering [29], computing and the design of factories [30], shops, offices, and hospitals [31]. e theory allows cloud system to be scaled optimally to guarantee the QoS for response time.…”

Taming tail latency for erasure-coded, distributee storage systems

“…t π-Optimization: Input t, S (18), (19), (20), (21), (22), (25) var. π S-Optimization: Input t, π (18), (19), (20), (21), (22), (23) var. S…”

“…This problem is challenging because (i) tail latency is significantly skewed by performance of the slowest storage nodes; (ii) a joint chunk scheduling problem needs to be solved on the fly to decide n-choose-k chunks/servers serving each file request; and (iii) the problem is further complicated by the dependency and interference of chunk access times of different files on shared storage servers. Toward this end, we make use of probabilistic scheduling proposed in [12,13,[21][22][23][24]. Upon the arrival of each file request, we randomly dispatch a batch of k chunk requests to k-out-of-n storage nodes selected with some predetermined probabilities.…”