A Survey of Data Center Network Architectures By Obasuyi Edokpolor

Slides:



Advertisements
Similar presentations
The Case for Enterprise Ready Virtual Private Clouds Timothy Wood, Alexandre Gerber *, K.K. Ramakrishnan *, Jacobus van der Merwe *, and Prashant Shenoy.
Advertisements

CSCI 465 D ata Communications and Networks Lecture 20 Martin van Bommel CSCI 465 Data Communications & Networks 1.
1 Exploring Efficient and Scalable Multicast Routing in Future Data Center Networks Dan Li, Jiangwei Yu, Junbiao Yu, Jianping Wu Tsinghua University Presented.
A Scalable, Commodity Data Center Network Architecture.
Mr. Mark Welton.  Three-tiered Architecture  Collapsed core – no distribution  Collapsed core – no distribution or access.
Chapter 1: Hierarchical Network Design
Network Support for Cloud Services Lixin Gao, UMass Amherst.
LAN Overview (part 2) CSE 3213 Fall April 2017.
The Future of GovNeTT – Where are we Heading?. GovNeTT 2.0 Current State Obscure Governance Framework Design is Difficult to Evolve to Changing Needs.
Network Aware Resource Allocation in Distributed Clouds.
Routing & Architecture
By Allan Joe Network Components and Terms. Contents Hubs Bridges Switches On-line &Off-line Bibliography.
LAN Switching and Wireless – Chapter 1
LAN Switching and Wireless – Chapter 1 Vilina Hutter, Instructor
© 2008 Cisco Systems, Inc. All rights reserved.Cisco ConfidentialPresentation_ID 1 Chapter 1: Introduction to Scaling Networks Scaling Networks.
IBM T. J. Watson Research © 2004 IBM Corporation On Scalable Storage Area Network(SAN) Fabric Design Algorithm Bong-Jun Ko (Columbia University) Kang-Won.
Networking Components Starla Wachsmann. COMPUTER NETWORKING COMPONETS Today’s wireless and enterprise networks are more complex than ever, delivering.
Data Communications and Networking CSCS 311 Lecture 4 Amjad Hussain Zahid.
1 Recommendations Now that 40 GbE has been adopted as part of the 802.3ba Task Force, there is a need to consider inter-switch links applications at 40.
Dual Centric Data Center Network Architectures DAWEI LI, JIE WU (TEMPLE UNIVERSITY) ZHIYONG LIU, AND FA ZHANG (CHINESE ACADEMY OF SCIENCES) ICPP 2015.
Interconnect Networks Basics. Generic parallel/distributed system architecture On-chip interconnects (manycore processor) Off-chip interconnects (clusters.
Data Communications and Networks Chapter 9 – Distributed Systems ICT-BVF8.1- Data Communications and Network Trainer: Dr. Abbes Sebihi.
Multi-objective Topology Synthesis and FPGA Prototyping Framework of Application Specific Network-on-Chip m Akram Ben Ahmed Xinyu LI, Omar Hammami.
Performance Evaluation of Parallel Algorithms on a Computational Grid Environment Simona Blandino 1, Salvatore Cavalieri 2 1 Consorzio COMETA, 2 Faculty.
© 2008 Cisco Systems, Inc. All rights reserved.Cisco ConfidentialPresentation_ID 1 Chapter 1: Hierarchical Network Design Connecting Networks.
Data Centers and Cloud Computing 1. 2 Data Centers 3.
CLOUD COMPUTING Presented to Graduate Students Mechanical Engineering Dr. John P. Abraham Professor, Computer Engineering UTPA.
“Your application performance is only as good as your network” (4)
Data Center Architectures
Warehouse Scaled Computers
INTERCONNECTION NETWORK
DENS: Data Center Energy-Efficient Network-Aware Scheduling
GreenCloud: A Packet-level Simulator of Energy-aware Cloud Computing Data Centers Dzmitry Kliazovich, Pascal Bouvry, Yury Audzevich, and Samee Ullah Khan.
Yiting Xia, T. S. Eugene Ng Rice University
Unit 3 Virtualization.
CIS 700-5: The Design and Implementation of Cloud Networks
Chapter 1: Exploring the Network
IMPROVING OF WIRELESS MESH NETWORKS.
Lecture 2: Cloud Computing
Network Overview.
Overview: Cloud Datacenters
Data Center Network Architectures
IOT Critical Impact on DC Design
Chuanxiong Guo, et al, Microsoft Research Asia, SIGCOMM 2008
What is Cloud Computing - How cloud computing help your Business?
3. Internetworking (part 1)
Principles of Network Applications
Chapter 15: Distributed Structures
Lecture 1: Network Operating Systems (NOS)
Network Configurations
Elastic Provisioning In Virtual Private Clouds
Virtualization, Cloud Computing and Big Data
The Improvement of PaaS Platform ZENG Shu-Qing, Xu Jie-Bin 2010 First International Conference on Networking and Distributed Computing SQUARE.
NTHU CS5421 Cloud Computing
Chapter 16: Distributed System Structures
Dr. John P. Abraham Professor, Computer Engineering UTPA
Module 5 - Switches CCNA 3 version 3.0.
ECE 4450:427/527 - Computer Networks Spring 2017
NTHU CS5421 Cloud Computing
Distributed computing deals with hardware
Computer Networking A computer network, often simply referred to as a network, is a collection of computers and devices connected by communications channels.
Computer Networking A computer network, often simply referred to as a network, is a collection of computers and devices connected by communications channels.
Internet and Web Simple client-server model
Data Center Architectures
Cloud-Enabling Technology
Applied Information Technology 2016 Rhys greaney
2019/5/13 A Weighted ECMP Load Balancing Scheme for Data Centers Using P4 Switches Presenter:Hung-Yen Wang Authors:Peng Wang, George Trimponias, Hong Xu,
Comm. And Networking By: Sam & parker.
Elmo Muhammad Shahbaz Lalith Suresh, Jennifer Rexford, Nick Feamster,
Presentation transcript:

A Survey of Data Center Network Architectures By Obasuyi Edokpolor Final Presentation A Survey of Data Center Network Architectures By Obasuyi Edokpolor

What is A Data Center? According to Wikipedia: A data center is a facility used to house computer systems and associated components, such as telecommunications and storage systems According to Google: A large group of networked computer servers typically used by organizations for the remote storage, processing, or distribution of large amounts of data.

Data Centers Infrastructure design has recently received significant interest due to the growing importance of data centers in supporting & sustaining the rapidly growing internet-based applications, which includes: Search Engines -> Google, Bing Video Content Hosting and Distribution -> Youtube, Netflix Social Networking -> Facebook, Twitter Large-Scale Computation -> Data Mining, Bioinformatics

Microsoft Live Online Services is supported by Chicago based data center which is one of the largest data center ever built, spanning 700,000 square feet Microsoft Chicago Data Center in Northlake, Illinois

Data Center Network Topology Due to the wide range of this survey, I will be focusing on Network Topology In other to make internet application work efficiently, it is imperative that data center are cost-effective and efficient. Most data centers use Ethernet switches to interconnect the servers, there are still many different ways to implement the interconnections, leading to different data center network topologies, each of these different topologies is characterized by different resource requirement, aiming to bring enhancement to the performance of data centers. The topology of the network interconnecting the servers, has a significant impart on the agility and configurability of the data center infrastructure to respond to changing application demands and service requirement.

Types of Network Topology According to Yang, network topology can be divided into two parts: Fixed Flexible Fixed topology : Is when a network topology cannot be modified after it has been deployed. Flexible Topology: Is when a network topology can be modified even after deployment Fixed Topology can be divided into 2 parts: Tree Recursive

Tree Topology There are 3 types of tree topology, which is: Basic Tree Fat Tree Clos Network Basic Tree : Consist of either 2 or 3 levels of switches/ routers, with the servers as leaf. 2-level of switches contains (a) Core Tier (b) Edge Tier 3-level of switches consist of (a) Core Tier (b) Aggregation Tier (c) Edge Tier

In a basic topology, the higher-tier switches need to support data communication among a large number of servers. This switches with higher performance and reliability are required in these tiers. The number of servers in a tree architecture is limited by number of ports on the switches.

Basic Tree

Fat Tree Fat Tree: Is like a basic tree based on a complete binary tree. Each n-port switches in the edge-tier is connected to n/2 servers. The remaining n/2 ports are connected to n/2 switches in the aggregation-level. The n/2 aggregation- level switches and the servers connected to the edge switch form a basic cell of fat tree, which is called a pod.

Fat Tree

Clos Network Is a multi-tier tree. When deployed in data center networks, a clos network usually consist of 3 levels of switches. The ToR switches directly connected to servers, the aggregation switches connected to the ToR switches, and the intermediate switches connected to the aggregation switches. These 3 level of switches are termed “input”, “middle” and “output” switches.

For example if each of the switches has n ports, there will be n aggregation switches and n/2 intermediate switches. There is exactly one link between each intermediate switches and each aggregation switch. The remaining n/2 ports on each aggregation switch is connected to n/2 different ToR switches. Each of the ToR switches is connected to 2 different aggregation switches, and remaining ports on the ToR

Recursive Topology It can scale-up by inserting more levels of switches, while each server is connected to only one of the bottom level switches. It is like a multi-tier tree-based topologies. However, recursive topologies use lower level structure of cells to build higher level structures, and the servers in the recursive topologies maybe connected to switches of different level or even other servers.

There are multiple network ports on the servers of recursive topologies, making them significantly different from the tree-based topologies. Graph are preferred to represent recursive architecture rather than rooted tree.

Type of Recursive Topology There are two types of recursive topology: DCell BCube Dcell : The basic element of a DCell, which is called DCello, consist of n servers and one n-port switch. Each server in a DCello is connected to the switch in the same DCello. Let Dcellk be a k-level Dcell. 1st construct DCell1 from several Dcell0. Each DCell1 has n+1 DCello, and each server of every DCellos in a DCell1 is connected to a server in another DCello, respectively As a result, the DCellos are connected to each other, with exactly one link between every pair of DCellos.

BCube: Is a recursive topology specially designed for shipping container based modular data centers. Building a data center cluster in a 20 and 40 foot shipping container makes it highly portable. As demand change at different data centers, the whole cluster can be readily moved. While deployment time is considerably shorter, the disadvantage of this environment is that due to operational and space constraints, once deployed in the field, it is difficult to service the cluster. The basic element of BCube is like that of DCello; n servers connected to one n-port switch. The only difference is how they scale up . BCube makes use of more switches when constructing higher level architecture. BCube1 contains n BCubeos and n extra switches

Flexible Topology Many researchers have considered using optical switching technology to construct Data Center Networks. Besides offering high bandwidth, optical networks have significant flexibility of reconfiguring the topology during operation. There are 3 types of Flexible topology c-Through Helios OSA

Bibliography A Survey of Data Center Network Architectures by Yang Liu https://web.njit.edu/~alexg/courses/cs345/solutions/index.html http://slideplayer.com/slide/4900969/