Stratos: A Network-Aware Orchestration Layer for Middleboxes in the Cloud Aditya Akella, Aaron Gember, Anand Krishnamurthy, Saul St. John University of.

Slides:



Advertisements
Similar presentations
Virtual Network Diagnosis as a Service Wenfei Wu (UW-Madison) Guohui Wang (Facebook) Aditya Akella (UW-Madison) Anees Shaikh (IBM System Networking)
Advertisements

Virtual Switching Without a Hypervisor for a More Secure Cloud Xin Jin Princeton University Joint work with Eric Keller(UPenn) and Jennifer Rexford(Princeton)
Big Data + SDN SDN Abstractions. The Story Thus Far Different types of traffic in clusters Background Traffic – Bulk transfers – Control messages Active.
Aaron Gember-Jacobson, Chaithan Prakash, Raajay Viswanathan, Robert Grandl, Junaid Khalid, Sourav Das, Aditya Akella 1 OpenNF: Enabling Innovation in Network.
Anand Krishnamurthy, Shoban P. Chandrabose and Aaron Gember-Jackobson 1 Pratyaastha: An Efficient Elastic Distributed SDN Control Plane.
SDN + Storage.
VCRIB: Virtual Cloud Rule Information Base Masoud Moshref, Minlan Yu, Abhishek Sharma, Ramesh Govindan HotCloud 2012.
Connect communicate collaborate GN3plus What the network should do for clouds? Christos Argyropoulos National Technical University of Athens (NTUA) Institute.
SLA-Oriented Resource Provisioning for Cloud Computing
Sharing Cloud Networks Lucian Popa, Gautam Kumar, Mosharaf Chowdhury Arvind Krishnamurthy, Sylvia Ratnasamy, Ion Stoica UC Berkeley.
Nanxi Kang Princeton University
Brocade VDX 6746 switch module for Hitachi Cb500
ECOS: Leveraging Software-Defined Networks to Support Mobile Application Offloading Aaron Gember, Christopher Dragga, Aditya Akella University of Wisconsin-Madison.
Towards Virtual Routers as a Service 6th GI/ITG KuVS Workshop on “Future Internet” November 22, 2010 Hannover Zdravko Bozakov.
Course Name- CSc 8320 Advanced Operating Systems Instructor- Dr. Yanqing Zhang Presented By- Sunny Shakya Latest AOS techniques, applications and future.
Scalable and Crash-Tolerant Load Balancing based on Switch Migration
Automatic Run-time Adaptation in Virtual Execution Environments Ananth I. Sundararaj Advisor: Peter A. Dinda Prescience Lab Department of Computer Science.
COMS E Cloud Computing and Data Center Networking Sambit Sahu
An Optimization Problem in Adaptive Virtual Environments Ananth I. Sundararaj Manan Sanghi Jack R. Lange Peter A. Dinda Prescience Lab Department of Computer.
SDN Scalability Issues
Multipath Protocol for Delay-Sensitive Traffic Jennifer Rexford Princeton University Joint work with Umar Javed, Martin Suchara, and Jiayue He
Jennifer Rexford Princeton University MW 11:00am-12:20pm Data-Center Traffic Management COS 597E: Software Defined Networking.
IOFlow: A Software-defined Storage Architecture Eno Thereska, Hitesh Ballani, Greg O’Shea, Thomas Karagiannis, Antony Rowstron, Tom Talpey, Richard Black,
1 Algorithms for Bandwidth Efficient Multicast Routing in Multi-channel Multi-radio Wireless Mesh Networks Hoang Lan Nguyen and Uyen Trang Nguyen Presenter:
Toward Software-Defined Middlebox Networking Aaron Gember, Prathmesh Prabhu, Zainab Ghadiyali, Aditya Akella University of Wisconsin-Madison 1.
DRFQ: Multi-Resource Fair Queueing for Packet Processing Ali Ghodsi 1,3, Vyas Sekar 2, Matei Zaharia 1, Ion Stoica 1 1 UC Berkeley, 2 Intel ISTC/Stony.
Network Sharing Issues Lecture 15 Aditya Akella. Is this the biggest problem in cloud resource allocation? Why? Why not? How does the problem differ wrt.
E-STAB: Energy-Efficient Scheduling for Cloud Computing Applications with Traffic Load Balancing Dzmitry KliazovichUniversity of Luxembourg, Luxembourg.
Network Aware Resource Allocation in Distributed Clouds.
DRFQ: Multi-Resource Fair Queueing for Packet Processing Ali Ghodsi 1,3, Vyas Sekar 2, Matei Zaharia 1, Ion Stoica 1 1 UC Berkeley, 2 Intel ISTC/Stony.
Improving Network I/O Virtualization for Cloud Computing.
SDN Dev Group, Week 3 Aaron GemberAditya Akella University of Wisconsin-Madison 1 Floodlight Controller; Application Wishlist.
Autonomic SLA-driven Provisioning for Cloud Applications Nicolas Bonvin, Thanasis Papaioannou, Karl Aberer Presented by Ismail Alan.
LAN Switching and Wireless – Chapter 1
CloudNaaS: A Cloud Networking Platform for Enterprise Applications Theophilus Benson*, Aditya Akella*, Anees Shaikh +, Sambit Sahu + (*University of Wisconsin,
Tag Switching Architecture Overview Qingfeng Zhuge Fangxia Li Xin Jiang.
Palette: Distributing Tables in Software-Defined Networks Yossi Kanizo (Technion, Israel) Joint work with Isaac Keslassy (Technion, Israel) and David Hay.
Advanced Resource Sharing in the Cloud Eiji Kawai NICT.
Software Defined Networks for Dynamic Datacenter and Cloud Environments.
Aaron Gember, Theophilus Benson, Aditya Akella University of Wisconsin-Madison.
Extending OVN Forwarding Pipeline Topology-based Service Injection
Resource Allocation in Network Virtualization Jie Wu Computer and Information Sciences Temple University.
Data Center Load Balancing T Seminar Kristian Hartikainen Aalto University, Helsinki, Finland
Piotr Srebrny 1.  Problem statement  Packet caching  Thesis claims  Contributions  Related works  Critical review of claims  Conclusions  Future.
Network Virtualization Sandip Chakraborty. In routing table we keep both the next hop IP (gateway) as well as the default interface. Why do we require.
Slide 1/20 "PerfSight: Performance Diagnosis for Software Dataplanes." Wu, Wenfei, Keqiang He, and Aditya Akella ACM ICM, Presented by: Ayush Patwari.
NEWS: Network Function Virtualization Enablement within SDN Data Plane.
Univ. of TehranIntroduction to Computer Network1 An Introduction to Computer Networks University of Tehran Dept. of EE and Computer Engineering By: Dr.
Logically Centralized? State Distribution Trade-offs in Software Defined Networks.
6.888 Lecture 6: Network Performance Isolation Mohammad Alizadeh Spring
Slide 1/12 Network Function Virtualization and its Dependability Challenges Relevant papers: 1.Gember-Jacobson, Aaron, Raajay Viswanathan, Chaithan Prakash,
@projectcalico Sponsored by Simple, Secure, Scalable networking for the virtualized datacentre UKNOF 33 Ed 19 th January 2016.
Chen Qian, Xin Li University of Kentucky
Yiting Xia, T. S. Eugene Ng Rice University
Xin Li, Chen Qian University of Kentucky
Optimizing Distributed Actor Systems for Dynamic Interactive Services
CIS 700-5: The Design and Implementation of Cloud Networks
Lecture 2: Cloud Computing
University of Maryland College Park
The DPIaaS Controller Prototype
P4P : Provider Portal for (P2P) Applications Haiyong Xie, Y
Heitor Moraes, Marcos Vieira, Italo Cunha, Dorgival Guedes
Network Anti-Spoofing with SDN Data plane Authors:Yehuda Afek et al.
Hydra: Leveraging Functional Slicing for Efficient Distributed SDN Controllers Yiyang Chang, Ashkan Rezaei, Balajee Vamanan, Jahangir Hasan, Sanjay Rao.
of Dynamic NFV-Policies
Storage Virtualization
Internet and Web Simple client-server model
Cloud Computing Architecture
Specialized Cloud Architectures
Towards Predictable Datacenter Networks
Presentation transcript:

Stratos: A Network-Aware Orchestration Layer for Middleboxes in the Cloud Aditya Akella, Aaron Gember, Anand Krishnamurthy, Saul St. John University of Wisconsin-Madison 1

Today’s cloud offerings Compute and storage are first-class entities – Rich management interfaces – Easy elasticity What about network services (middleboxes)? 2 [Sherry et al., SIGCOMM 2012] Limited cloud-provided middleboxes Third party virtual middlebox images

VM Difficult to deploy complex functionality Difficult to manage Difficult to cost-effectively scale Insufficient support for middleboxes 3 VM App B App B App B App B App A App A VM

Stratos Network-aware orchestration layer for middleboxes in clouds Elevates network services to a first-class entity Exports a logical view (middlebox chains) to tenants Performs application-specific, network-aware scaling Minimizes network effects => ↑ utilization and ↓ cost Requires no knowledge of/changes to middleboxes Driven completely by software (leverages SDN) Key to Stratos: network awareness 4

Rack ARack B Why network awareness – I Scale based on resource consumption App Low CPU Usage Congested link Request backlog Ignoring the network  insufficient scaling 5

Why network awareness – II 6 App Scaling doesn’t help Request backlog Place VMs without regard to the network Ignoring the network  ineffective scaling

Rack A Rack B Equally divide traffic among middleboxes Why network awareness – III 1/2 of traffic traverses inter-rack link Ignoring network  over-utilized network Network bottlenecks  spurious scaling 7

Stratos Controller Stratos architecture 8 VM Manager PlacementFlow Distribution A A B B Software SDN Switches Scaling

Stratos scaling Based on end-to-end application performance – Implicitly compute- and network- aware Occurs at the granularity of chains Triggers – Scale up: ↑ chain-traversal latency OR existence of unserved demand – Scale down: ↓ request throughput AND ≈ constant chain-traversal latency 9

Scaling trials on a chain If ↓ Latency OR ↓ demand backlog: Keep and try another Else: Discard and move on Fallback: scale all Also supports scale down and multiple chains App Server Stratos scaling (single chain) ms400 ms395 ms

Stratos Controller Stratos architecture 11 VM Manager Flow Distribution A A B B Software SDN Switches ScalingPlacement

Initial placement 12 A A B B B B A A k=1; done = false While (!done): Identify k partitions //min-K-cut If partitions can be accommodated: done = true Else: k++ k=1; done = false While (!done): Identify k partitions //min-K-cut If partitions can be accommodated: done = true Else: k++

Scaled instance placement 13 A A B B B B A A If space with input/output VMs: Co-locate in same rack Else Foreach rack i bwc i = b/w consumed if use rack i Pick rack with min bwc i If space with input/output VMs: Co-locate in same rack Else Foreach rack i bwc i = b/w consumed if use rack i Pick rack with min bwc i

Stratos Controller Stratos architecture 14 VM Manager A A B B Software SDN Switches ScalingPlacementFlow Distribution

Goal: minimize network effects Triggers – Scaling (tenant-specific) – Periodically (all tenants) Network-aware flow distribution 15 Rack A Rack B 1 / 6 of traffic (instead of 1 / 2 ) Linear Program Input: tenant chain, incoming traffic volume, traffic ratios, placement Minimize: overall “cost” (aggregate traffic traversing inter-rack links) Subject to: ≈ equal load; coverage Linear Program Input: tenant chain, incoming traffic volume, traffic ratios, placement Minimize: overall “cost” (aggregate traffic traversing inter-rack links) Subject to: ≈ equal load; coverage

Floodlight Xen Implementation 16 dom0 domU Open vSwitch eth0 Stratos Controller

Implementation – tagging Controller assigns tags to each flow – Tag identifies path through specific instances – Weighted round-robin assignment of tags to flows Packets tagged (use DSCP bits) at ingress switch “Interior” switches forward based on tag 17 Open vSwitch App Tag Packets Forward based on tag

Evaluation: Placement & Distribution 18 Spurious scaling Unmet demand Spurious scaling (not pronounced) Unmet demand

Evaluation: Scaling 19 Scaling/Placement/Distribution Aware – ours Thresh - CPU Aware – ours Rand - random Aware – ours Uni - uniform A A 2X fewer Unmet demand

Stratos Summary Network-aware orchestration layer for middleboxes in clouds Makes middleboxes first-class citizens Minimizes network interactions Maximizes efficiency for tenants and providers Driven by software-defined networking 20