Advanced Computer Networks. Datacenter TCP
|
|
- Shauna Davidson
- 6 years ago
- Views:
Transcription
1 Advanced Computer Networks Datacenter TCP Patrick Stuedi, Qin Yin, Timothy Roscoe Spring Semester Oriana Riva, Department of Computer Science ETH Zürich
2 Last week Datacenter Fabric Portland VL2 Facebook Altoona 2
3 Today Datacenter traffic analysis TCP incast Fine-grained TCP retransmissions Data Center TCP (DCTCP) Deadline-aware TCP Multipath TCP 3
4 Why bother about TCP? Applications require reliable end-to-end data transfer Typically implemented using TCP Problem: TCP designed for the Internet, performance in other networks often suboptimal Example: High bandwidth/delay product networks Two options: Improve TCP for other networks: SnoopTCP for wireless, DCTCP for datacenter Use other protocol instead How easy is it replace TCP? Easy if application is written from scratch Possible if we control both ends of the communication (like in a datacenter) Difficult if we are not controlling both ends: protocol often dictated by the server (e.g., HTTP/TCP webserver) 4
5 The Partition/Aggegrate Pattern 5
6 How does search work? Picasso Picasso
7 How Does Search Work? Picasso Picasso source: stanford CS244
8 How Does Search Work?
9 Partition / Aggregate Foundation of many large web applications Web-search, social networks, ad selection Example: Facebook Aggregators: web servers Workers: Memcached servers Memcached source: stanford CS244
10 Measureing Traffic of a Datacenter 10
11 Two interesting studies The Nature of Data Center Traffice: Measurement & Analycis, IMC, 2009 Looks at MapReduce Datacenters Measuring traffic on servers (intercept socket calls) Network Traffic Characteristics of Data Centers in the Wild, IMC 2010 Wide range of data centers (Universities and Commercial) Measurements using SNMP counters at switches and packet traces Measurement difficulties: make sure measurement technique does not affect performance (e.g., CPU utilization) All datacenters 2-tiered or 3-tiered topologies 11
12 Who talks to whom? Two patterns dominate Most traffic stays within the rack Scatter/gather 12
13 Intra-rack versus extra-rack communciation Extra-Rack Inter-Rack EDU1 EDU2 EDU3 PRV1 PRV2 CLD1 CLD2 CLD3 CLD4 CLD5 Clouds: most traffic stays within rack Other DCs: > 50 % of the traffic leaves the rack 13
14 Are the links busy? Many links experience occasional congestion 86% of the links observe congestion of at least 10s 15% of the links observe congestion of at least 100s 14
15 Link utilization across DC layers Utilization: core > aggr > edge Cloud DCs: core links often loaded > 70% (e.g., CLD5) 15
16 Congestion length Contiguous Duration of >70% link utilization (seconds) Most periods of congestion tend to be short lived 86% of all congestion events last at less than 10s 16
17 Flow size 65% of Flows are < 1MB 1 CDF 0.8 Flow Size Total Bytes Flow Size (Bytes) % of Bytes from Flows > 1MB 10 8
18 Datacenter Workloads Mice & Elephants Short messages (50KB-1MB) (query, coordination, control state) Delay sensitive Large Flows (1MB-100MB) (data update, backup) Throughput sensitive 18
19 Measurement Take Away Cloud DCs: Large parts of the traffic stays within a rack University DCs: 50% of the traffic leaves the rack Link utilization low at edge and aggregation layer Core most utilized, > 70% in Cloud DCs Congestion of flows typically last a few seconds Most flows are small (< 1MB) But it is the large flows that make up for most of the bytes Workload composed of short delay-sensitive message, and large throughput-sensitive flows 19
20 Refresh: TCP Congestion Control 20
21 Remember: TCP congestion control Congestion control got added to TCP to in attempt to reduce congestion inside the network Must rely on indirect measures of congestion Implemented at the sender Src Dest Attempts to reduce buffer overflow inside the network 21
22 Remember: TCP Slow Start Congestion window (CW) Number of bytes in TCP that can be transmitted without waiting for the ACK (CW always smaller than receiver window, flow ctrl) Initially set to 1 TCP segment SSThresh Initially set to 64 KB TCP congestion control: After all ACKs corresponding to one CW have been received (typically after one RTT), the window is doubled - slow start (actually quite fast) If CW >= SSThresh increase CW by 1 TCP segment after all ACKs corresponding to one CW have been received - linear increase (congestion avoidance) On a timeout: Set SSThresh to half of the current CW, then set CW back to 1K 22
23 Example: Slow start 23
24 Incast 24
25 Problem: TCP Incast Worker 1 Synchronized fan-in congestion: Caused by Partition/Aggregate. Worker 2 Worker 3 Aggregator RTOmin = 300 ms Worker 4 TCP timeout 25
26 TCP Incast (2) Incast event measured in a production environment Request forwarded in over 0.8 ms (800 microseconds) All but one response returning in 12.4 ms Retransmission after RTO: 300 ms 26
27 Throughput collapse Collapse! Cause of throughput collapse: coarse-grained TCP timeouts 27
28 Approach: Fine-grained timeouts Roundtrip timeout typically set based on measured RTT+X, with RTO >= RTO_min Two problems: Most Linux TCP implementations do not measure RTT as fine granular as needed for datacenters (Linux jiffies updated per second) RTO_min typically too large Idea: Reduce RTO_min Measure RTT using high-resolution timers in us granularity 28
29 Lower timeout helps microsecond TCP + no minrto 1ms minrto Unmodified TCP (200ms minrto) more servers High throughput for up to 47 servers 29
30 Data Center TCP (DCTCP) 30
31 Queue Buildup Queue Buildup Sender 1 Large flows buildup queues: Increase latency for short flows. Receiver Sender 2 drop
32 Approach: Datacenter TCP Datacenter TCP (DCTCP): Mark packets in switches using Explicit Congestion Notification (ECN) if they experience congestion Scale the TCP window down proportionally to the number of packets with ECN bit set ECN Marks TCP DCTCP Cut window by 50% Cut window by 40% Cut window by 50% Cut window by 5% 32
33 DCTCP Algorithm Switch-side: Mark packets if Queue length > K using ECN bit Receiver-side: Echo bit back to sender with delayed ACKs Sender-side: Maintain running average a of fraction of packets marked (value of 'a' between 0 and 1) - a = (1-g)*a + g*f F: fraction of packets marked in last window 0 > g < 1 : weight given to new samples - a close to 0 means low congestion - a close to 1 means high congestion Window decrease in case of ECN-marked ACK: w = w x (1-a/2) 33
34 DCTCP in Action DCTCP achieves full throughput (not shown in Figure) while taking up a very small footprint in the switch 34
35 Deadline-aware TCP 35
36 User-facing online services Partition/aggregate workflow Application SLAs Cascading SLAs Network SLAs Flow deadlines A flow is useful if and only if it satisfies its deadline 36
37 Limitations of fair sharing (1) Flows f1 and f2 get a fair share of bandwidth Flow f1 misses its deadline (incomplete response to user) Case for unfair sharing: Flows get bandwith in accordance to their deadlines Deadline awareness ensures both flows satisfy deadlines 37
38 Limitations of fair sharing (1) Flows f1 and f2 get a fair share of bandwidth Flow f1 misses its deadline (incomplete response to user) Case for unfair sharing: Flows get bandwith in accordance to their deadlines Deadline awareness ensures both flows satisfy deadlines 38
39 Limitations of fair sharing (2) Insufficient bandwidth to satisfy all deadlines With fair share, all flows miss the deadline (empty response) Case for flow quenching: With deadline awareness, one flow can be quenched All other flows make their deadline 39
40 Limitations of fair sharing (2) Insufficient bandwidth to satisfy all deadlines With fair share, all flows miss the deadline (empty response) Case for flow quenching: With deadline awareness, one flow can be quenched All other flows make their deadline 40
41 D3: Deadline driven delivery Main idea: make the network aware of flow deadlines Prioritize flows based on deadlines Key insight: Rate required to satisfy a flow deadline: r = s / d s: flow size d: deadline 41
42 How it works (1) Application exposes (s,d) Desired rate r = s / d Routers allocate rates (α) based on traffic load Sending rate for next RTT : sr = min(α1,α2) s: flow size d: deadline RRQ: rate request α: allocated rate One of the packets contains and updated RRQ based on the remaining flow size and the deadline 42
43 How it works (2) Application exposes (s,d) Desired rate r = s / d Routers allocate rates (α) based on traffic load Sending rate for next RTT : sr = min(α1,α2) s: flow size d: deadline RRQ: rate request α: allocated rate One of the packets contains and updated RRQ based on the remaining flow size and the deadline 43
44 Flow microbenchmarks Experiment: multiple workers sending traffic to aggregator (within a single rack) How many workers can be supported while satisfying deadlines? D3 can support roughly twice as many workers than RCP (~DCTCP) while satisfying application deadlines 44
45 Flow quenching Terminate useless flows when: Desired rate exceeds link capacity Deadline has expired 45
46 D3 Summary Traditional TCP flow-sharing leads flows missing deadlines D3 allocated rates at switches based on the deadlines of the flows D3 can support many more flows with deadlines than TCP 46
47 Limitations of D3 Needs router support User-space PC-based implementation for paper Violates end-to-end argument Greedy rate allocation leads to priority inversion and later to missed deadlines 47
48 Deadline-aware Datacenter TCP 2 (D TCP) Key idea: Vary sending rate based on both deadline and extent of congestion Built on top of DCTCP Per-flow state at end-hosts (not routers/switches) Reactive: senders react to congestion No knowledge of other flows 48
49 D2TCP Congestion Avoidance (1) Remember: DTCP congestion control TCP window: w = w x (1-a/2) Running average of marked packets: a = (1-g)*a + g*f (a ~ 0: low congestion, a ~ 1: high congestion) D2TCP extends DTCP to integrate deadline Deadline factor d: larger d implies closer deadline Penalty function: p = ad TCP window: w = w * (1 p/2) if p > 0 w=w+1 if p = 0 Note: a = 0 => p = 0 => w = w +1 (similar to TCP) a = 1 => p = 1 => w = w/2 (similar to TCP) 49
50 D2TCP Congestion Avoidance (2) d < 1 for far deadline flows => p large => shrink window d > 1 for near deadline flows => p small => retain window d = 1 for long lived flows => DTCP behavior Near deadline flows back off less while far deadline flows back off mor 50
51 How to determine d? d = Tc / D Tc: time needed for a flow to complete under deadlineagnostic congestion behavior (based on the current window w) D: remaining time until deadline expires Flow is on track: TC D => d 1 Flow is about to miss deadline: TC > D => d > 1 Flow is ahead of deadline TC < D => d < 1 51
52 D2TCP versus DCTCP Flow sizes: 150MB, 220MB, 350MB, 500MB Flow deadlines: 1000ms, 1500ms, 2500ms, 4000ms DTCP: all flows get same b/w irrespective of deadline D2TCP: Near deadline flows get more bandwidth 52
53 Multipath TCP 53
54 Modern datacenters provide many parallel paths Traditional topologies are treebased Poor performance Not fault-tolerant Shift towards multipath topologies FatTree (Portland, VL2) BCube 54
55 Modern datacenters provide many parallel paths Traditional topologies are treebased Poor performance Not fault-tolerant Shift towards multipath topologies FatTree (Portland, VL2) BCube How to effectively use all the bandwidth in a multipath topology? 55
56 Equal-cost multipath routing (ECMP) ECMP Multipath routing strategy that splits traffic over multiple paths for loadbalancing Why not just round-robin packets? Reordering: triple TCP ACKs, TCP fast retransmit, reduced TCP window Different RTT per path Different MTUs per path Path selection via hashing #buckets = #outgoing links Hash network information (src address/port, dst address/port, protocol type) to select outgoing link: preserves flow affinity 56
57 ECMP in Fat Tree Network Hash flows to different paths 57
58 ECMP collision Multiple flows may hash to same path 58
59 Limitations of ECMP ECMP may not utilize the links uniformly Many flows hashed to same link ECMP is static No knowledge about current traffic on a link 59
60 Limitations of ECMP (2) 8192 node fat tree toplogy Random traffic pattern: Flow: every host chooses a random destination No destination is used twice Figure: flows ranked according to their throughput 60
61 Limitations of ECMP (2) 8192 node fat tree toplogy Random traffic pattern: Flow: every host chooses a random destination Vast majority of flows does not achieve No destination is used twice more than 50% of the throughput possible Figure: flows ranked according to their throughput 61
62 Multipath TCP (MPTCP) Instead of using one path for each flow, many random paths Don't worry about collisions Just don't send (much) traffic on colliding path 62
63 Multipath TCP (2) MPTCP establishes mutiple subflows on different paths With many random paths, MPTCP will find at least one good unloaded path and moves most of its traffic on that path Drop-in replacement Improves... Bandwidth (aggregation) Fairness Robustness 63
64 MPTCP in Fat Tree Network Vast majority of flows using MPTCP achieve % throughput 64
65 MTCP: Connection Management 65
66 MTCP: Connection Management 66
67 MTCP: Connection Management 67
68 MTCP: Connection Management 68
69 MTCP: Connection Management client learns about additional interfaces of server 69
70 MTCP: Connection Management Works in data centers, problem when using MPTCP across the Internet: 6% of access networks 70remove unknown options
71 MTCP: Connection Management 71
72 MTCP: Connection Management 72
73 MTCP: Connection Management Subflows can be between different interfaces or between the same pair of IP addresses but different ports 73
74 MTCP: Connection Management 74
75 MTCP: Connection Management 75
76 MTCP: Connection Management MPTCP relies on ECMP to hash different subflows to different paths 76
77 MTCP: Sending Data MPTCP stripes TCP across the subflows Additional TCP options allow the receiver to reconstruct the received data in the original order 77
78 MTCP: Sending Data ACK1 ACK3 ACK2 ACK4 MPTCP stripes TCP across the subflows When used over the Internet, middleboxes may drop ACKs of78unseen data packets
79 MPTCP Congestion Control Each path runs its own congestion control, to detect and respond to the congestion it sees But link congestion control parameters, so as to move traffic away from the congested paths 79
80 How many subflows are needed? 80
81 Next week Flow control Infiniband Converged enhanced Ethernet 81
Advanced Computer Networks. Datacenter TCP
Advanced Computer Networks 263 3501 00 Datacenter TCP Spring Semester 2017 1 Oriana Riva, Department of Computer Science ETH Zürich Today Problems with TCP in the Data Center TCP Incast TPC timeouts Improvements
More informationAdvanced Computer Networks. Flow Control
Advanced Computer Networks 263 3501 00 Flow Control Patrick Stuedi Spring Semester 2017 1 Oriana Riva, Department of Computer Science ETH Zürich Last week TCP in Datacenters Avoid incast problem - Reduce
More informationBetter Never than Late: Meeting Deadlines in Datacenter Networks
Better Never than Late: Meeting Deadlines in Datacenter Networks Christo Wilson, Hitesh Ballani, Thomas Karagiannis, Ant Rowstron Microsoft Research, Cambridge User-facing online services Two common underlying
More informationTCP Incast problem Existing proposals
TCP Incast problem & Existing proposals Outline The TCP Incast problem Existing proposals to TCP Incast deadline-agnostic Deadline-Aware Datacenter TCP deadline-aware Picasso Art is TLA 1. Deadline = 250ms
More informationAdvanced Computer Networks. Wireless TCP
Advanced Computer Networks 263 3501 00 Wireless TCP Patrick Stuedi Spring Semester 2014 1 Oriana Riva, Department of Computer Science ETH Zürich Outline Last week: Today: Cellular Networks Mobile IP Wireless
More informationData Center TCP (DCTCP)
Data Center TCP (DCTCP) Mohammad Alizadeh, Albert Greenberg, David A. Maltz, Jitendra Padhye Parveen Patel, Balaji Prabhakar, Sudipta Sengupta, Murari Sridharan Microsoft Research Stanford University 1
More informationLecture 15: Datacenter TCP"
Lecture 15: Datacenter TCP" CSE 222A: Computer Communication Networks Alex C. Snoeren Thanks: Mohammad Alizadeh Lecture 15 Overview" Datacenter workload discussion DC-TCP Overview 2 Datacenter Review"
More informationData Center TCP (DCTCP)
Data Center Packet Transport Data Center TCP (DCTCP) Mohammad Alizadeh, Albert Greenberg, David A. Maltz, Jitendra Padhye Parveen Patel, Balaji Prabhakar, Sudipta Sengupta, Murari Sridharan Cloud computing
More informationAdvanced Computer Networks. Flow Control
Advanced Computer Networks 263 3501 00 Flow Control Patrick Stuedi, Qin Yin, Timothy Roscoe Spring Semester 2015 Oriana Riva, Department of Computer Science ETH Zürich 1 Today Flow Control Store-and-forward,
More informationCloud e Datacenter Networking
Cloud e Datacenter Networking Università degli Studi di Napoli Federico II Dipartimento di Ingegneria Elettrica e delle Tecnologie dell Informazione DIETI Laurea Magistrale in Ingegneria Informatica Prof.
More informationData Center TCP(DCTCP)
Data Center TCP(DCTCP) Mohammad Alizadeh * +, Albert Greenberg *, David A. Maltz *, Jitendra Padhye *, Parveen Patel *, Balaji Prabhakar +, Sudipta Sengupta *, Murari Sridharan * * + Microsoft Research
More informationComputer Networks Spring 2017 Homework 2 Due by 3/2/2017, 10:30am
15-744 Computer Networks Spring 2017 Homework 2 Due by 3/2/2017, 10:30am (please submit through e-mail to zhuoc@cs.cmu.edu and srini@cs.cmu.edu) Name: A Congestion Control 1. At time t, a TCP connection
More informationUtilizing Datacenter Networks: Centralized or Distributed Solutions?
Utilizing Datacenter Networks: Centralized or Distributed Solutions? Costin Raiciu Department of Computer Science University Politehnica of Bucharest We ve gotten used to great applications Enabling Such
More informationRecap. TCP connection setup/teardown Sliding window, flow control Retransmission timeouts Fairness, max-min fairness AIMD achieves max-min fairness
Recap TCP connection setup/teardown Sliding window, flow control Retransmission timeouts Fairness, max-min fairness AIMD achieves max-min fairness 81 Feedback Signals Several possible signals, with different
More informationEXPERIENCES EVALUATING DCTCP. Lawrence Brakmo, Boris Burkov, Greg Leclercq and Murat Mugan Facebook
EXPERIENCES EVALUATING DCTCP Lawrence Brakmo, Boris Burkov, Greg Leclercq and Murat Mugan Facebook INTRODUCTION Standard TCP congestion control, which only reacts to packet losses has many problems Can
More informationDeTail Reducing the Tail of Flow Completion Times in Datacenter Networks. David Zats, Tathagata Das, Prashanth Mohan, Dhruba Borthakur, Randy Katz
DeTail Reducing the Tail of Flow Completion Times in Datacenter Networks David Zats, Tathagata Das, Prashanth Mohan, Dhruba Borthakur, Randy Katz 1 A Typical Facebook Page Modern pages have many components
More informationPacket Scheduling in Data Centers. Lecture 17, Computer Networks (198:552)
Packet Scheduling in Data Centers Lecture 17, Computer Networks (198:552) Datacenter transport Goal: Complete flows quickly / meet deadlines Short flows (e.g., query, coordination) Large flows (e.g., data
More informationTransport layer issues
Transport layer issues Dmitrij Lagutin, dlagutin@cc.hut.fi T-79.5401 Special Course in Mobility Management: Ad hoc networks, 28.3.2007 Contents Issues in designing a transport layer protocol for ad hoc
More informationCommunication Networks
Communication Networks Spring 2018 Laurent Vanbever nsg.ee.ethz.ch ETH Zürich (D-ITET) April 30 2018 Materials inspired from Scott Shenker & Jennifer Rexford Last week on Communication Networks We started
More informationCS519: Computer Networks. Lecture 5, Part 4: Mar 29, 2004 Transport: TCP congestion control
: Computer Networks Lecture 5, Part 4: Mar 29, 2004 Transport: TCP congestion control TCP performance We ve seen how TCP the protocol works Sequencing, receive window, connection setup and teardown And
More informationData Center TCP (DCTCP)
Data Center TCP (DCTCP) Mohammad Alizadeh, Albert Greenberg, David A. Maltz, Jitendra Padhye Parveen Patel, Balaji Prabhakar, Sudipta Sengupta, Murari Sridharan Stanford University MicrosoD Research Case
More informationLecture 21: Congestion Control" CSE 123: Computer Networks Alex C. Snoeren
Lecture 21: Congestion Control" CSE 123: Computer Networks Alex C. Snoeren Lecture 21 Overview" How fast should a sending host transmit data? Not to fast, not to slow, just right Should not be faster than
More information15-744: Computer Networking. Data Center Networking II
15-744: Computer Networking Data Center Networking II Overview Data Center Topology Scheduling Data Center Packet Scheduling 2 Current solutions for increasing data center network bandwidth FatTree BCube
More informationCross-Layer Flow and Congestion Control for Datacenter Networks
Cross-Layer Flow and Congestion Control for Datacenter Networks Andreea Simona Anghel, Robert Birke, Daniel Crisan and Mitch Gusat IBM Research GmbH, Zürich Research Laboratory Outline Motivation CEE impact
More informationLecture 14: Congestion Control"
Lecture 14: Congestion Control" CSE 222A: Computer Communication Networks George Porter Thanks: Amin Vahdat, Dina Katabi and Alex C. Snoeren Lecture 14 Overview" TCP congestion control review Dukkipati
More informationCSE/EE 461 Lecture 16 TCP Congestion Control. TCP Congestion Control
CSE/EE Lecture TCP Congestion Control Tom Anderson tom@cs.washington.edu Peterson, Chapter TCP Congestion Control Goal: efficiently and fairly allocate network bandwidth Robust RTT estimation Additive
More informationData center Networking: New advances and Challenges (Ethernet) Anupam Jagdish Chomal Principal Software Engineer DellEMC Isilon
Data center Networking: New advances and Challenges (Ethernet) Anupam Jagdish Chomal Principal Software Engineer DellEMC Isilon Bitcoin mining Contd Main reason for bitcoin mines at Iceland is the natural
More informationTCP Congestion Control
6.033, Spring 2014 TCP Congestion Control Dina Katabi & Sam Madden nms.csail.mit.edu/~dina Sharing the Internet How do you manage resources in a huge system like the Internet, where users with different
More informationGot Loss? Get zovn! Daniel Crisan, Robert Birke, Gilles Cressier, Cyriel Minkenberg, and Mitch Gusat. ACM SIGCOMM 2013, August, Hong Kong, China
Got Loss? Get zovn! Daniel Crisan, Robert Birke, Gilles Cressier, Cyriel Minkenberg, and Mitch Gusat ACM SIGCOMM 2013, 12-16 August, Hong Kong, China Virtualized Server 1 Application Performance in Virtualized
More informationCongestion control in TCP
Congestion control in TCP If the transport entities on many machines send too many packets into the network too quickly, the network will become congested, with performance degraded as packets are delayed
More informationTransport Protocols for Data Center Communication. Evisa Tsolakou Supervisor: Prof. Jörg Ott Advisor: Lect. Pasi Sarolahti
Transport Protocols for Data Center Communication Evisa Tsolakou Supervisor: Prof. Jörg Ott Advisor: Lect. Pasi Sarolahti Contents Motivation and Objectives Methodology Data Centers and Data Center Networks
More informationTCP. The TCP Protocol. TCP Header Format. TCP Flow Control. TCP Congestion Control Datacenter TCP
The TCP Protocol TCP TCP Congestion Control Datacenter TCP Frame format Connection management Flow control TCP reliable data transfer Congestion control TCP Header Format TCP Flow Control HL 32 bits Source
More informationCongestion Control in Datacenters. Ahmed Saeed
Congestion Control in Datacenters Ahmed Saeed What is a Datacenter? Tens of thousands of machines in the same building (or adjacent buildings) Hundreds of switches connecting all machines What is a Datacenter?
More informationComputer Networking
15-441 Computer Networking Lecture 17 TCP Performance & Future Eric Anderson Fall 2013 www.cs.cmu.edu/~prs/15-441-f13 Outline TCP modeling TCP details 2 TCP Performance Can TCP saturate a link? Congestion
More informationTCP. CSU CS557, Spring 2018 Instructor: Lorenzo De Carli (Slides by Christos Papadopoulos, remixed by Lorenzo De Carli)
TCP CSU CS557, Spring 2018 Instructor: Lorenzo De Carli (Slides by Christos Papadopoulos, remixed by Lorenzo De Carli) 1 Sources Fall and Stevens, TCP/IP Illustrated Vol. 1, 2nd edition Congestion Avoidance
More informationImpact of transmission errors on TCP performance. Outline. Random Errors
Impact of transmission errors on TCP performance 1 Outline Impact of transmission errors on TCP performance Approaches to improve TCP performance Classification Discussion of selected approaches 2 Random
More informationDIBS: Just-in-time congestion mitigation for Data Centers
DIBS: Just-in-time congestion mitigation for Data Centers Kyriakos Zarifis, Rui Miao, Matt Calder, Ethan Katz-Bassett, Minlan Yu, Jitendra Padhye University of Southern California Microsoft Research Summary
More informationCongestion Control. Tom Anderson
Congestion Control Tom Anderson Bandwidth Allocation How do we efficiently share network resources among billions of hosts? Congestion control Sending too fast causes packet loss inside network -> retransmissions
More informationLecture 14: Congestion Control"
Lecture 14: Congestion Control" CSE 222A: Computer Communication Networks Alex C. Snoeren Thanks: Amin Vahdat, Dina Katabi Lecture 14 Overview" TCP congestion control review XCP Overview 2 Congestion Control
More informationCS268: Beyond TCP Congestion Control
TCP Problems CS68: Beyond TCP Congestion Control Ion Stoica February 9, 004 When TCP congestion control was originally designed in 1988: - Key applications: FTP, E-mail - Maximum link bandwidth: 10Mb/s
More informationCongestion. Can t sustain input rate > output rate Issues: - Avoid congestion - Control congestion - Prioritize who gets limited resources
Congestion Source 1 Source 2 10-Mbps Ethernet 100-Mbps FDDI Router 1.5-Mbps T1 link Destination Can t sustain input rate > output rate Issues: - Avoid congestion - Control congestion - Prioritize who gets
More informationAdvanced Computer Networks. RDMA, Network Virtualization
Advanced Computer Networks 263 3501 00 RDMA, Network Virtualization Patrick Stuedi Spring Semester 2013 Oriana Riva, Department of Computer Science ETH Zürich Last Week Scaling Layer 2 Portland VL2 TCP
More informationCS 344/444 Computer Network Fundamentals Final Exam Solutions Spring 2007
CS 344/444 Computer Network Fundamentals Final Exam Solutions Spring 2007 Question 344 Points 444 Points Score 1 10 10 2 10 10 3 20 20 4 20 10 5 20 20 6 20 10 7-20 Total: 100 100 Instructions: 1. Question
More informationCSE 123A Computer Networks
CSE 123A Computer Networks Winter 2005 Lecture 14 Congestion Control Some images courtesy David Wetherall Animations by Nick McKeown and Guido Appenzeller The bad news and the good news The bad news: new
More informationComputer Network Fundamentals Spring Week 10 Congestion Control Andreas Terzis
Computer Network Fundamentals Spring 2008 Week 10 Congestion Control Andreas Terzis Outline Congestion Control TCP Congestion Control CS 344/Spring08 2 What We Know We know: How to process packets in a
More informationT Computer Networks II Data center networks
23.9.20 T-110.5116 Computer Networks II Data center networks 1.12.2012 Matti Siekkinen (Sources: S. Kandula et al.: The Nature of Datacenter: measurements & analysis, A. Greenberg: Networking The Cloud,
More informationAttaining the Promise and Avoiding the Pitfalls of TCP in the Datacenter. Glenn Judd Morgan Stanley
Attaining the Promise and Avoiding the Pitfalls of TCP in the Datacenter Glenn Judd Morgan Stanley 1 Introduction Datacenter computing pervasive Beyond the Internet services domain BigData, Grid Computing,
More informationMicro load balancing in data centers with DRILL
Micro load balancing in data centers with DRILL Soudeh Ghorbani (UIUC) Brighten Godfrey (UIUC) Yashar Ganjali (University of Toronto) Amin Firoozshahian (Intel) Where should the load balancing functionality
More informationMPTCP: Design and Deployment. Day 11
MPTCP: Design and Deployment Day 11 Use of Multipath TCP in ios 7 Multipath TCP in ios 7 Primary TCP connection over WiFi Backup TCP connection over cellular data Enables fail-over Improves performance
More informationPage 1. Review: Internet Protocol Stack. Transport Layer Services. Design Issue EEC173B/ECS152C. Review: TCP
EEC7B/ECS5C Review: Internet Protocol Stack Review: TCP Application Telnet FTP HTTP Transport Network Link Physical bits on wire TCP LAN IP UDP Packet radio Transport Layer Services Design Issue Underlying
More informationComputer Networks (Fall 2011) Homework 2
5-744 Computer Networks (Fall 20) Homework 2 Name: Due: Oct. 2th, 20, 3:00PM (in class) Andrew ID: October 2, 20 A Short Questions. Which of the following is true about modern high-speed routers? A. A
More informationFast Retransmit. Problem: coarsegrain. timeouts lead to idle periods Fast retransmit: use duplicate ACKs to trigger retransmission
Fast Retransmit Problem: coarsegrain TCP timeouts lead to idle periods Fast retransmit: use duplicate ACKs to trigger retransmission Packet 1 Packet 2 Packet 3 Packet 4 Packet 5 Packet 6 Sender Receiver
More informationRouting Domains in Data Centre Networks. Morteza Kheirkhah. Informatics Department University of Sussex. Multi-Service Networks July 2011
Routing Domains in Data Centre Networks Morteza Kheirkhah Informatics Department University of Sussex Multi-Service Networks July 2011 What is a Data Centre? Large-scale Data Centres (DC) consist of tens
More informationCS644 Advanced Networks
What we know so far CS644 Advanced Networks Lecture 6 Beyond TCP Congestion Control Andreas Terzis TCP Congestion control based on AIMD window adjustment [Jac88] Saved Internet from congestion collapse
More information15-744: Computer Networking TCP
15-744: Computer Networking TCP Congestion Control Congestion Control Assigned Reading [Jacobson and Karels] Congestion Avoidance and Control [TFRC] Equation-Based Congestion Control for Unicast Applications
More informationTransmission Control Protocol. ITS 413 Internet Technologies and Applications
Transmission Control Protocol ITS 413 Internet Technologies and Applications Contents Overview of TCP (Review) TCP and Congestion Control The Causes of Congestion Approaches to Congestion Control TCP Congestion
More informationInternet Networking recitation #10 TCP New Reno Vs. Reno
recitation #0 TCP New Reno Vs. Reno Spring Semester 200, Dept. of Computer Science, Technion 2 Introduction Packet Loss Management TCP Reno (RFC 258) can manage a loss of at most one packet from a single
More informationPage 1. Review: Internet Protocol Stack. Transport Layer Services EEC173B/ECS152C. Review: TCP. Transport Layer: Connectionless Service
EEC7B/ECS5C Review: Internet Protocol Stack Review: TCP Application Telnet FTP HTTP Transport Network Link Physical bits on wire TCP LAN IP UDP Packet radio Do you remember the various mechanisms we have
More informationNetwork Management & Monitoring
Network Management & Monitoring Network Delay These materials are licensed under the Creative Commons Attribution-Noncommercial 3.0 Unported license (http://creativecommons.org/licenses/by-nc/3.0/) End-to-end
More informationImproving the Robustness of TCP to Non-Congestion Events
Improving the Robustness of TCP to Non-Congestion Events Presented by : Sally Floyd floyd@acm.org For the Authors: Sumitha Bhandarkar A. L. Narasimha Reddy {sumitha,reddy}@ee.tamu.edu Problem Statement
More informationEECS 122, Lecture 19. Reliable Delivery. An Example. Improving over Stop & Wait. Picture of Go-back-n/Sliding Window. Send Window Maintenance
EECS 122, Lecture 19 Today s Topics: More on Reliable Delivery Round-Trip Timing Flow Control Intro to Congestion Control Kevin Fall, kfall@cs cs.berkeley.eduedu Reliable Delivery Stop and Wait simple
More informationComputer Networks. Course Reference Model. Topic. Congestion What s the hold up? Nature of Congestion. Nature of Congestion 1/5/2015.
Course Reference Model Computer Networks 7 Application Provides functions needed by users Zhang, Xinyu Fall 204 4 Transport Provides end-to-end delivery 3 Network Sends packets over multiple links School
More informationOverview. TCP & router queuing Computer Networking. TCP details. Workloads. TCP Performance. TCP Performance. Lecture 10 TCP & Routers
Overview 15-441 Computer Networking TCP & router queuing Lecture 10 TCP & Routers TCP details Workloads Lecture 10: 09-30-2002 2 TCP Performance TCP Performance Can TCP saturate a link? Congestion control
More information8. TCP Congestion Control
8. TCP Congestion Control 1 TCP Congestion Control Slow-start increase Multiplicative decrease Congestion avoidance Measurement of variation Exponential timer backoff 2002 Yanghee Choi 2 Congestion Control
More informationCongestion Avoidance and Control. Rohan Tabish and Zane Ma
Congestion Avoidance and Control Rohan Tabish and Zane Ma TCP is self-clocking Self-clocking systems should be robust Congestion collapse Internet had first of what became a series of congestion collapses
More informationLet It Flow Resilient Asymmetric Load Balancing with Flowlet Switching
Let It Flow Resilient Asymmetric Load Balancing with Flowlet Switching Erico Vanini*, Rong Pan*, Mohammad Alizadeh, Parvin Taheri*, Tom Edsall* * Load Balancing in Data Centers Multi-rooted tree 1000s
More information6.033 Spring 2015 Lecture #11: Transport Layer Congestion Control Hari Balakrishnan Scribed by Qian Long
6.033 Spring 2015 Lecture #11: Transport Layer Congestion Control Hari Balakrishnan Scribed by Qian Long Please read Chapter 19 of the 6.02 book for background, especially on acknowledgments (ACKs), timers,
More informationMASSACHUSETTS INSTITUTE OF TECHNOLOGY Computer Systems Engineering: Spring Quiz I
Department of Electrical Engineering and Computer Science MASSACHUSETTS INSTITUTE OF TECHNOLOGY 6.033 Computer Systems Engineering: Spring 2016 Quiz I There are 15 questions and 13 pages in this quiz booklet.
More informationChapter 3 outline. 3.5 Connection-oriented transport: TCP. 3.6 Principles of congestion control 3.7 TCP congestion control
Chapter 3 outline 3.1 Transport-layer services 3.2 Multiplexing and demultiplexing 3.3 Connectionless transport: UDP 3.4 Principles of reliable data transfer 3.5 Connection-oriented transport: TCP segment
More informationCS 421: COMPUTER NETWORKS SPRING FINAL May 24, minutes. Name: Student No: TOT
CS 421: COMPUTER NETWORKS SPRING 2012 FINAL May 24, 2012 150 minutes Name: Student No: Show all your work very clearly. Partial credits will only be given if you carefully state your answer with a reasonable
More informationTCP over Wireless. Protocols and Networks Hadassah College Spring 2018 Wireless Dr. Martin Land 1
TCP over Wireless Protocols and Networks Hadassah College Spring 218 Wireless Dr. Martin Land 1 Classic TCP-Reno Ideal operation in-flight segments = cwnd (send cwnd without stopping) Cumulative ACK for
More informationTCP improvements for Data Center Networks
TCP improvements for Data Center Networks Tanmoy Das and rishna M. Sivalingam Department of Computer Science and Engineering, Indian Institute of Technology Madras, Chennai 636, India Emails: {tanmoy.justu@gmail.com,
More informationGood Ideas So Far Computer Networking. Outline. Sequence Numbers (reminder) TCP flow control. Congestion sources and collapse
Good Ideas So Far 15-441 Computer Networking Lecture 17 TCP & Congestion Control Flow control Stop & wait Parallel stop & wait Sliding window Loss recovery Timeouts Acknowledgement-driven recovery (selective
More informationOutline Computer Networking. TCP slow start. TCP modeling. TCP details AIMD. Congestion Avoidance. Lecture 18 TCP Performance Peter Steenkiste
Outline 15-441 Computer Networking Lecture 18 TCP Performance Peter Steenkiste Fall 2010 www.cs.cmu.edu/~prs/15-441-f10 TCP congestion avoidance TCP slow start TCP modeling TCP details 2 AIMD Distributed,
More informationCongestion Control End Hosts. CSE 561 Lecture 7, Spring David Wetherall. How fast should the sender transmit data?
Congestion Control End Hosts CSE 51 Lecture 7, Spring. David Wetherall Today s question How fast should the sender transmit data? Not tooslow Not toofast Just right Should not be faster than the receiver
More informationReliable Transport II: TCP and Congestion Control
Reliable Transport II: TCP and Congestion Control Stefano Vissicchio UCL Computer Science COMP0023 Recap: Last Lecture Transport Concepts Layering context Transport goals Transport mechanisms and design
More informationCS419: Computer Networks. Lecture 10, Part 3: Apr 13, 2005 Transport: TCP performance
: Computer Networks Lecture 10, Part 3: Apr 13, 2005 Transport: TCP performance TCP performance We ve seen how TCP the protocol works But there are a lot of tricks required to make it work well Indeed,
More informationCMPE 257: Wireless and Mobile Networking
CMPE 257: Wireless and Mobile Networking Katia Obraczka Computer Engineering UCSC Baskin Engineering Lecture 10 CMPE 257 Spring'15 1 Student Presentations Schedule May 21: Sam and Anuj May 26: Larissa
More informationCS321: Computer Networks Congestion Control in TCP
CS321: Computer Networks Congestion Control in TCP Dr. Manas Khatua Assistant Professor Dept. of CSE IIT Jodhpur E-mail: manaskhatua@iitj.ac.in Causes and Cost of Congestion Scenario-1: Two Senders, a
More informationOn the Efficacy of Fine-Grained Traffic Splitting Protocols in Data Center Networks
Purdue University Purdue e-pubs Department of Computer Science Technical Reports Department of Computer Science 2011 On the Efficacy of Fine-Grained Traffic Splitting Protocols in Data Center Networks
More informationimage 3.8 KB Figure 1.6: Example Web Page
image. KB image 1 KB Figure 1.: Example Web Page and is buffered at a router, it must wait for all previously queued packets to be transmitted first. The longer the queue (i.e., the more packets in the
More informationTCP Strategies. Keepalive Timer. implementations do not have it as it is occasionally regarded as controversial. between source and destination
Keepalive Timer! Yet another timer in TCP is the keepalive! This one is not required, and some implementations do not have it as it is occasionally regarded as controversial! When a TCP connection is idle
More informationT Computer Networks II Data center networks
T-110.5116 Computer Networks II Data center networks 29.9.2014 Matti Siekkinen (Sources: S. Kandula et al.: The Nature of Datacenter: measurements & analysis, A. Greenberg: Networking The Cloud, M. Alizadeh
More informationCSE 473 Introduction to Computer Networks. Midterm Exam Review
CSE 473 Introduction to Computer Networks Midterm Exam Review John DeHart 10/12/2015 1. (10 points). A user in Chicago, connected to the internet via a 100 Mb/s (b=bits) connection retrieves a 250 KB (B=bytes)
More informationTCP Congestion Control
TCP Congestion Control What is Congestion The number of packets transmitted on the network is greater than the capacity of the network Causes router buffers (finite size) to fill up packets start getting
More informationTCP Congestion Control
What is Congestion TCP Congestion Control The number of packets transmitted on the network is greater than the capacity of the network Causes router buffers (finite size) to fill up packets start getting
More informationADVANCED COMPUTER NETWORKS
ADVANCED COMPUTER NETWORKS Congestion Control and Avoidance 1 Lecture-6 Instructor : Mazhar Hussain CONGESTION CONTROL When one part of the subnet (e.g. one or more routers in an area) becomes overloaded,
More informationNetworking for Data Acquisition Systems. Fabrice Le Goff - 14/02/ ISOTDAQ
Networking for Data Acquisition Systems Fabrice Le Goff - 14/02/2018 - ISOTDAQ Outline Generalities The OSI Model Ethernet and Local Area Networks IP and Routing TCP, UDP and Transport Efficiency Networking
More information15-441: Computer Networks Homework 3
15-441: Computer Networks Homework 3 Assigned: Oct 29, 2013 Due: Nov 12, 2013 1:30 PM in class Name: Andrew ID: 1 TCP 1. Suppose an established TCP connection exists between sockets A and B. A third party,
More informationCS244 Advanced Topics in Computer Networks Midterm Exam Monday, May 2, 2016 OPEN BOOK, OPEN NOTES, INTERNET OFF
CS244 Advanced Topics in Computer Networks Midterm Exam Monday, May 2, 2016 OPEN BOOK, OPEN NOTES, INTERNET OFF Your Name: Answers SUNet ID: root @stanford.edu In accordance with both the letter and the
More informationEquation-Based Congestion Control for Unicast Applications. Outline. Introduction. But don t we need TCP? TFRC Goals
Equation-Based Congestion Control for Unicast Applications Sally Floyd, Mark Handley AT&T Center for Internet Research (ACIRI) Jitendra Padhye Umass Amherst Jorg Widmer International Computer Science Institute
More informationChapter III: Transport Layer
Chapter III: Transport Layer UG3 Computer Communications & Networks (COMN) Mahesh Marina mahesh@ed.ac.uk Slides thanks to Myungjin Lee and copyright of Kurose and Ross Principles of congestion control
More informationCongestion Control in TCP
Congestion Control in TCP Antonio Carzaniga Faculty of Informatics University of Lugano November 11, 2014 Outline Intro to congestion control Input rate vs. output throughput Congestion window Congestion
More informationcs/ee 143 Communication Networks
cs/ee 143 Communication Networks Chapter 4 Transport Text: Walrand & Parakh, 2010 Steven Low CMS, EE, Caltech Recap: Internet overview Some basic mechanisms n Packet switching n Addressing n Routing o
More informationCongestion Control. Queuing Discipline Reacting to Congestion Avoiding Congestion. Issues
Congestion Control Outline Queuing Discipline Reacting to Congestion Avoiding Congestion Issues Two sides of the same coin pre-allocate resources to avoid congestion (e.g. telephone networks) control congestion
More informationLecture 15: TCP over wireless networks. Mythili Vutukuru CS 653 Spring 2014 March 13, Thursday
Lecture 15: TCP over wireless networks Mythili Vutukuru CS 653 Spring 2014 March 13, Thursday TCP - recap Transport layer TCP is the dominant protocol TCP provides in-order reliable byte stream abstraction
More informationCongestion Control In The Internet Part 2: How it is implemented in TCP. JY Le Boudec 2014
1 Congestion Control In The Internet Part 2: How it is implemented in TCP JY Le Boudec 2014 Contents 1. Congestion control in TCP 2. The fairness of TCP 3. The loss throughput formula 4. Explicit Congestion
More informationCongestion Control In The Internet Part 2: How it is implemented in TCP. JY Le Boudec 2015
1 Congestion Control In The Internet Part 2: How it is implemented in TCP JY Le Boudec 2015 Contents 1. Congestion control in TCP 2. The fairness of TCP 3. The loss throughput formula 4. Explicit Congestion
More informationIIP Wireless. Presentation Outline
IIP Wireless Improving Internet Protocols for Wireless Links Markku Kojo Department of Computer Science www.cs cs.helsinki.fi/research/.fi/research/iwtcp/ 1 Presentation Outline Project Project Summary
More informationCongestion Control in TCP
Congestion Control in TCP Antonio Carzaniga Faculty of Informatics University of Lugano May 6, 2005 Outline Intro to congestion control Input rate vs. output throughput Congestion window Congestion avoidance
More information