IMPROVING DATACENTER PERFORMANCE AND ROBUSTNESS WITH MULTIPATH TCP Costin Raiciu†, Sebastien Barre‡, Christopher Pluntke†, Adam Greenhalgh†, Damon Wischik†,

Slides:



Advertisements
Similar presentations
Data Center Networking with Multipath TCP
Advertisements

Improving Datacenter Performance and Robustness with Multipath TCP
Algorithms Chapter 15 Dynamic Programming - Rod
Improving Datacenter Performance and Robustness with Multipath TCP Costin Raiciu, Sebastien Barre, Christopher Pluntke, Adam Greenhalgh, Damon Wischik,
Utilizing Datacenter Networks: Dealing with Flow Collisions Costin Raiciu Department of Computer Science University Politehnica of Bucharest.
Risk and Asset Allocation CH3 Modeling the market 3.1~ 黃欣裕.
Pricing and Power Control in a Multicell Wireless Data Network Po Yu Chen October, 2001 IEEE Journal on Select Areas in Communications.
第 七 章 開放式問句與探究.
Chapter 10 馬可夫鏈 緒言 如果讀者仔細觀察日常生活中所發生的 諸多事件,必然會發現有些事件的未來 發展或演變與該事件現階段的狀況全然 無關,這種事件稱為獨立試行過程 (process of independent trials) ;而另一些 事件則會受到該事件現階段的狀況影響。
第七章 抽樣與抽樣分配 蒐集統計資料最常見的方式是抽查。這 牽涉到兩個問題: 抽出的樣本是否具有代表性?是否能反應出母體的特徵?
: A-Sequence 星級 : ★★☆☆☆ 題組: Online-judge.uva.es PROBLEM SET Volume CIX 題號: Problem D : A-Sequence 解題者:薛祖淵 解題日期: 2006 年 2 月 21 日 題意:一開始先輸入一個.
:Word Morphing ★★☆☆☆ 題組: Problem Set Archive with Online Judge 題號: 10508:word morphing 解題者:楊家豪 解題日期: 2006 年 5 月 21 日 題意: 第一行給你兩個正整數, 第一個代表下面會出現幾個字串,
Teacher : Ing-Jer Huang TA : Chien-Hung Chen 2015/6/2 Course Embedded Systems : Principles and Implementations Weekly Preview Question CH 1.4~CH /10/03.
Section 1.2 Describing Distributions with Numbers 用數字描述分配.
Teacher : Ing-Jer Huang TA : Chien-Hung Chen 2015/6/3 Course Embedded Systems : Principles and Implementations Weekly Preview Question CH3.5 ~ CH /10/31.
3Com Switch 4500 切VLAN教學.
期末專題 - 吊人頭遊戲 第 35 組 組員 : 電機系 49841XXXX XXX 電機系 49841OOOO OOO.
What is static?. Static? 靜態 ? class Test { static int staticX; int instanceX; public Test(int var1, int var2) { this.staticX = var1; this.instanceX =
指導教授:陳淑媛 學生:李宗叡 李卿輔.  利用下列三種方法 (Edge Detection 、 Local Binary Pattern 、 Structured Local Edge Pattern) 來判斷是否為場景變換,以方便使用者來 找出所要的片段。
廣告故事學 --為商品注入生命的行銷陽謀. 從前,有一家半大不小的連鎖商店, 沒有什麼裝潢,也不具備設計感, 看起來不起眼,設的點也不多,更 不能刷卡,說實在的,怎麼看就是 破舊老陳的。但是,這家店的缺點 突然變成優點 …… 廣告故事 A-1.
Datacenter Network Topologies
交易的動機. 討論:  為什麼人要進行交易? 討論:  試想像你走進一間唱片店,付了$100 給店主,買下你最喜愛歌手的最新唱片。  誰會得益 ?  誰又會有所損失?
倫理準則:機密性. Confidentiality By: Angela Lo. 倫理準則:機密性. Confidentiality 醫護人員有更多的機會接觸病患的隱私。 隱私包括兩方面︰一是病患的身體,另一 是有關病患的機密的訊息。 醫護人員有更多的機會接觸病患的隱私。 隱私包括兩方面︰一是病患的身體,另一.
我的生涯 巫俊賢.
亂數產生器安全性評估 之統計測試 SEC HW7 姓名:翁玉芬 學號:
Review of Chapter 3 - 已學過的 rules( 回顧 )- 朝陽科技大學 資訊管理系 李麗華 教授.
:New Land ★★★★☆ 題組: Problem Set Archive with Online Judge 題號: 11871: New Land 解題者:施博修 解題日期: 2011 年 6 月 8 日 題意:國王有一個懶兒子,為了勞動兒子,他想了一個 辦法,令他在某天早上開始走路,直到太陽下山前,靠.
STAT0_sampling Random Sampling  母體: Finite population & Infinity population  由一大小為 N 的有限母體中抽出一樣本數為 n 的樣 本,若每一樣本被抽出的機率是一樣的,這樣本稱 為隨機樣本 (random sample)
基礎物理總論 基礎物理總論 熱力學與統計力學(三) Statistical Mechanics 東海大學物理系 施奇廷.
Monte Carlo Simulation Part.2 Metropolis Algorithm Dept. Phys. Tunghai Univ. Numerical Methods C. T. Shih.
Hint of Homework 4 jinnjy. Outline Hint of exercise 3.18.
2009fallStat_samplec.i.1 Chap10 Sampling distribution (review) 樣本必須是隨機樣本 (random sample) ,才能代表母體 Sample mean 是一隨機變數,隨著每一次抽出來的 樣本值不同,它的值也不同,但會有規律性 為了要知道估計的精確性,必需要知道樣本平均數.
Network Connections ★★★☆☆ 題組: Contest Archive with Online Judge 題號: Network Connections 解題者:蔡宗翰 解題日期: 2008 年 10 月 20 日 題意:給你電腦之間互相連線的狀況後,題.
Introduction to Java Programming Lecture 17 Abstract Classes & Interfaces.
: The largest Clique ★★★★☆ 題組: Contest Archive with Online Judge 題號: 11324: The largest Clique 解題者:李重儀 解題日期: 2008 年 11 月 24 日 題意: 簡單來說,給你一個 directed.
Matlab Assignment Due Assignment 兩個 matlab 程式 : Eigenface : Eigenvector 和 eigenvalue 的應用. Fractal : Affine transform( rotation, translation,
: Count DePrimes ★★★★☆ 題組: Contest Archive with Online Judge 題號: 11408: Count DePrimes 解題者:李育賢 解題日期: 2008 年 9 月 2 日 題意: 題目會給你二個數字 a,b( 2 ≦ a ≦ 5,000,000,a.
短缺,盈餘與均衡. 遊戲規則  老師想出售一些學生喜歡的小食。  老師首先講出價錢,有興趣買的請舉手。
: Multisets and Sequences ★★★★☆ 題組: Problem Set Archive with Online Judge 題號: 11023: Multisets and Sequences 解題者:葉貫中 解題日期: 2007 年 4 月 24 日 題意:在這個題目中,我們要定義.
公司加入市場的決定. 定義  平均成本 = 總成本 ÷ 生產數量 = 每一單位產量所耗的成本  平均固定成本 = 總固定成本 ÷ 生產數量  平均變動成本 = 總變動成本 ÷ 生產數量.
:Nuts for nuts..Nuts for nuts.. ★★★★☆ 題組: Problem Set Archive with Online Judge 題號: 10944:Nuts for nuts.. 解題者:楊家豪 解題日期: 2006 年 2 月 題意: 給定兩個正整數 x,y.
SPSS 分析簡介 何明洲 中山醫學大學心理系. 資料在 SPSS 上之排列 Between-subject design, one factor with three levels.
公用品.  該物品的數量不會因一人的消費而受到 影響,它可以同時地被多人享用。 角色分配  兩位同學當我的助手,負責:  其餘各人是投資者,每人擁有 $100 , 可以投資在兩種資產上。  記錄  計算  協助同學討論.
: Problem G e-Coins ★★★☆☆ 題組: Problem Set Archive with Online Judge 題號: 10306: Problem G e-Coins 解題者:陳瀅文 解題日期: 2006 年 5 月 2 日 題意:給定一個正整數 S (0
Section 4.2 Probability Models 機率模式. 由實驗看機率 實驗前先列出所有可能的實驗結果。 – 擲銅板:正面或反面。 – 擲骰子: 1~6 點。 – 擲骰子兩顆: (1,1),(1,2),(1,3),… 等 36 種。 決定每一個可能的實驗結果發生機率。 – 實驗後所有的實驗結果整理得到。
Teacher : Ing-Jer Huang TA : Chien-Hung Chen 2015/6/25 Course Embedded Systems : Principles and Implementations Weekly Preview Question CH 2.4~CH 2.6 &
演算法 8-1 最大數及最小數找法 8-2 排序 8-3 二元搜尋法.
Density and control Reproduction curve 劉耀仁. Density :濃度、密度 ► 1 號區是 High-Density 區、 2 號是 Middle-density 區、 3 號區是 Low- Density 區。 ► 計算反射率( R )或透射率( T )
-Antidifferentiation- Chapter 6 朝陽科技大學 資訊管理系 李麗華 教授.
逆向選擇和市場失調. 定義  資料不對稱 在交易其中,其中一方較對方有多些資料。  逆向選擇 出現在這個情況下,就是當買賣雙方隨意在 市場上交易,與比較主動交易者作交易為佳 。
: Flip Sort ★★☆☆☆ 題組: Problem Set Archive with Online Judge 題號: 10327: Flip Sort 解題者:歐子揚 解題日期: 2010 年 2 月 26 日 題意:在這個問題中使用一種排序方式 (Flip) ,意思就是 只能交換相鄰的.
845: Gas Station Numbers ★★★ 題組: Problem Set Archive with Online Judge 題號: 845: Gas Station Numbers. 解題者:張維珊 解題日期: 2006 年 2 月 題意: 將輸入的數字,經過重新排列組合或旋轉數字,得到比原先的數字大,
JAVA 程式設計與資料結構 第十六章 Hash Tables. Introduction Hash Tables 結構為一個 Array ,稱之為 Bucket array 。 如果想要新增一個物件,要根據這個物件的特性 將其加入 Hash Table 內。 Bucket Array 用 A 來代替,其.
2005/7 Linear system-1 The Linear Equation System and Eliminations.
連續隨機變數 連續變數:時間、分數、重量、……
Teacher : Ing-Jer Huang TA : Chien-Hung Chen 2015/6/30 Course Embedded Systems : Principles and Implementations Weekly Preview Question CH7.1~CH /12/26.
McGraw-Hill/Irwin © 2003 The McGraw-Hill Companies, Inc.,All Rights Reserved. 參 資料蒐集的方法.
協助孩子從閱讀學習英文 吳敏而 香港教育學院 家長的責任   安排學習空間   安排學習材料   安排學習時間   陪孩子   關懷孩子   鼓勵孩子.
: SAM I AM ★★★★☆ 題組: Contest Archive with Online Judge 題號: 11419: SAM I AM 解題者:李重儀 解題日期: 2008 年 9 月 11 日 題意: 簡單的說,就是一個長方形的廟裡面有敵人,然 後可以橫的方向開砲或縱向開砲,每次開砲可以.
:Count the Trees ★★★☆☆ 題組: Problem Set Archive with Online Judge 題號: 10007:Count the Trees 解題者:楊家豪 解題日期: 2006 年 3 月 題意: 給 n 個點, 每一個點有自己的 Label,
: Finding Paths in Grid ★★★★☆ 題組: Contest Archive with Online Judge 題號: 11486: Finding Paths in Grid 解題者:李重儀 解題日期: 2008 年 10 月 14 日 題意:給一個 7 個 column.
:Problem E.Stone Game ★★★☆☆ 題組: Problem Set Archive with Online Judge 題號: 10165: Problem E.Stone Game 解題者:李濟宇 解題日期: 2006 年 3 月 26 日 題意: Jack 與 Jim.
:Rings and Glue ★★☆☆☆ 題組: Problem Set Archive with Online Judge 題號: 10301: Rings and Glue 解題者:施博修 解題日期: 2011 年 5 月 18 日 題意:小約翰有了個大麻煩,他不小心將 rings.
幼兒行為觀察與記錄 第八章 事件取樣法.
: How many 0's? ★★★☆☆ 題組: Problem Set Archive with Online Judge 題號: 11038: How many 0’s? 解題者:楊鵬宇 解題日期: 2007 年 5 月 15 日 題意:寫下題目給的 m 與 n(m
1 Slide Slide 第 9 章 假設檢定 Part B ( ). 2 Slide Slide 第 9 章 假設檢定 Part B  9.5 母體比例  9.6 假設檢定與決策  9.7 計算型 II 錯誤的機率  9.8 在檢定母體平均數時決定樣本大小 第 9 章假設檢定.
1 柱體與錐體 1. 找出柱體與錐體的規則 2. 柱體的命名與特性 3. 柱體的展開圖 4. 錐體的命名與特性 5. 錐體的展開圖
Congestion control for Multipath TCP (MPTCP) Damon Wischik Costin Raiciu Adam Greenhalgh Mark Handley THE ROYAL SOCIETY.
Improving Datacenter Performance and Robustness with Multipath TCP
Improving Datacenter Performance and Robustness with Multipath TCP
Presentation transcript:

IMPROVING DATACENTER PERFORMANCE AND ROBUSTNESS WITH MULTIPATH TCP Costin Raiciu†, Sebastien Barre‡, Christopher Pluntke†, Adam Greenhalgh†, Damon Wischik†, Mark Handley†

Before Start  我想要先用我的話說一次這篇 Paper 想做的事情 :  這邊 Paper 的重點在於, single path 用在 TCP 上面 的時代已經過去了, multipath TCP 才是主流。  Multipath TCP 又主要有三種 topologies: fattree, VL2, Bcube 。  這篇 paper 又分析了在不同情況下 multipath 確實 比 single path 優秀的地方。還有一些影響 performance 的因素的分析。  最後是有沒有其他可能再發展的 topology

OUTLINE  Introduction  Data center networking  Topology.  Routing.  Path Selection.  Congestion Control.  Multipath TCP in summary  MPTCP in data center  Examples of Benefits  Analysis Influence of Topology Number of Subflows Influence of the Traffic Matrix  Evolving topologies with MPTCP  Analysis  Discussion about DHFT  Summary

Introduction  Traditional single path TCP has many significant problems, ill-suited for such network. (intra-data center traffic.)  Ex. Bottleneck, can’t scale up, congestion.  We propose using Multipath TCP as a replacement for TCP in such data centers.

Introduction  Traditional data center topology. …

Introduction  4 components to a data center architecture:  1. physical topology  2. routing  3. path selection  4. congestion control of traffic on the selected path.

OUTLINE  Introduction  Data center networking  Topology.  Routing.  Path Selection.  Congestion Control.  Multipath TCP in summary  MPTCP in data center  Examples of Benefits  Analysis Influence of Topology Number of Subflows Influence of the Traffic Matrix  Evolving topologies with MPTCP  Analysis  Discussion about DHFT  Summary

Data center networking---Topology  Traditional data center topology. …

Data center networking---Topology  That has many problems and can be improved a lot like we have discussed before.  So, fat-tree is one kind of solutions.

Fat Tree Topology [Fares et al., 2008; Clos, 1953] K=4 Aggregation Switches K Pods with K Switches each Racks of servers

Data center networking---Topology  And we also have VL2 and Bcube these two kinds of solutions.

BCube Topology [Guo et al, 2009] BCube (4,1)

Data center networking---Topology

VL2 Topology [Greenberg et al, 2009, Clos topology] 10Gbps 20 hosts 10Gbps …

Data center networking---Routing  We cannot expect the host itself to know which of these paths is the least loaded, so the routing system must spread traffic across these paths.  Use randomized load balancing, where each flow is assigned a random path from the set of possible paths.

Data center networking--- Path Selection  randomized load balancing 有他一定的問題,就 是有一定的機率某些 path 會有超大流量、同時 會有一些 path 沒什麼 ( 極低或零 ) 流量。  The use of a centralized flow scheduler has been proposed.

Multipath TCP: Congestion Control [NSDI, 2011]

Data center networking--- Congestion Control  Our hypothesis is that given sufficiently many randomly chosen paths, MPTCP will find at least one good unloaded path, and move most of its traffic that way. In so doing it will relieve congestion on links that got more than their fair share of ECMP balanced flows.

Multipath TCP in summary  By making the window increase depend on the total window size, subflows that have large windows increase faster than subflows with small windows.

OUTLINE  Introduction  Data center networking  Topology.  Routing.  Path Selection.  Congestion Control.  Multipath TCP in summary  MPTCP in data center  Examples of Benefits  Analysis Influence of Topology Number of Subflows Influence of the Traffic Matrix  Evolving topologies with MPTCP  Analysis  Discussion about DHFT  Summary

MPTCP in data center  Three main benefits:  1. Better aggregate throughput  2. Better fairness  3. Better robustness

MPTCP in data center  Depend on 4 requirements:  1. The congestion control scheme used.  2. The physical topology.  3. The traffic matrix generated by the applications.  4. The level of load in the network.

Examples of Benefits  Throughput:  Using MPTCP in a FatTree network, we can see the more subflows we have, the more throughput we have.  The reason is simple, n flows share one link, each flow need to achieves 1/n capacity.

Examples of Benefits

 Fairness:  Every host’s throughput is shown ranked in order of increasing throughput. Is is clear that not only did the utilization improve with MPTCP, but also the fairness improved.

Examples of Benefits  這是一個累積百分比的圖,可以看出 single TCP 有些 path perform 真的很差 ( 很閒的意思 ) ,而 MPTCP 則大幅改善此情況。

Analysis  Then, we have questions:  How many subflows are needed?  How does the topology affect results?  How does the traffic matrix affect results?

Analysis  We need to view in these aspects to answer the question:  Influence of Topology  Number of Subflows  Influence of the Traffic Matrix

Influence of Topology  We compare VL2 and Bcube.

Influence of Topology  Why Bcube is relative low ?  I guess, just guess, is because in Bcube, pathes between two hosts are not in same length which lead to worse flows management.

Number of Subflows  We were surprised that eight subflows were needed to achieve 90% throughput.  Why eight, and on what does this depend?

Number of Subflows  Results:

Influence of the Traffic Matrix

OUTLINE  Introduction  Data center networking  Topology.  Routing.  Path Selection.  Congestion Control.  Multipath TCP in summary  MPTCP in data center  Examples of Benefits  Analysis Influence of Topology Number of Subflows Influence of the Traffic Matrix  Evolving topologies with MPTCP  Analysis  Discussion about DHFT  Summary

Evolving topologies with MPTCP  In single homed topologies:  Hosts links are often bottlenecked.  ToR switches failure also lead to crash tens of hosts.

Evolving topologies with MPTCP  以 FatTree 為例子

Evolving topologies with MPTCP  So, how do we improved ?  There are two solutions: perfect switch and dual- homed FatTree.

Evolving topologies with MPTCP  Perfect switch:  Serves as a good control experiment, giving an upper bound on what any network core might provide using single links to the hosts.

Evolving topologies with MPTCP  Dual-homed FatTree (DHFT):  is to the host and four connect the links between the two layers of switches.  If we remove one port per host from the core and use it to connect the second interface on each server, the network requires the same number of switch ports.

Analysis  QUESTION:  Will perfect switch really improve performance ?

Analysis  Testing throughput over MPTCP, perfect switch TCP, normal TCP.

Analysis  QUESTION:  Will perfect switch really improve performance ?  ANSWER:  Yes, it is.

Analysis  QUESTION:  Dual-homed FatTree really work ?  Does it have any problem ? When may troubles occur ?

Analysis  Testing relative throughput over MPTCP, TCP DHFT, TCP perfect switch.

Analysis  QUESTION:  Dual-homed FatTree really work ?  Does it have any problem ? When may troubles occur ?  ANSWER:  Yes, it works only when load is low. When load is high, throughput will go down near to TCP.

Discussion about DHFT  DHFT’s worst case performance is 75% and best case is around 200%.  在流量不穩定的情況下, DHFT + MPTCP 有可能會有 較差的表現產生。  Beyond performance, DHFT improves robustness: any lowerpod switch failure does not cut-off an entire rack of servers.  DHFT is not optimal by any measure, but it shows that we can create topologies with better performance if we assume MPTCP is the transport protocol.

OUTLINE  Introduction  Data center networking  Topology.  Routing.  Path Selection.  Congestion Control.  Multipath TCP in summary  MPTCP in data center  Examples of Benefits  Analysis Influence of Topology Number of Subflows Influence of the Traffic Matrix  Evolving topologies with MPTCP  Analysis  Discussion about DHFT  Summary

Summary  “One flow, one path” thinking has constrained datacenter design  Collisions, unfairness, limited utilization  Multipath transport enables resource pooling in datacenter networks:  Improves throughput  Improves fairness  Improves robustness

Reference  conferences.sigcomm.org/sigcomm/2011/slides/s2 66.ppt ‎

 Q & (A)