Upload
angelina-greene
View
215
Download
2
Tags:
Embed Size (px)
Citation preview
Advanced Network Managementcourse#: 80240663
Spring 2015Instructor: Dan Pei
Teaching Assistant: Ruming Tang [email protected]
2
How to pronounce “Pei”?
Louvre MuseumI-M Pei: the famous architect who designed the glass pyramid
Just pronounce “Pei” as “Pay”
In fact, just call me “Dan”
3
About the Instructor• Associate Professor. Joined Tsinghua CS Department in December 2012, with
Government Endorsement (“ Recruitment Program of Global Talents”) Homepage: http://netman.cs.tsinghua.edu.cn/~peidan
• Previously at AT&T Research, worked on Management of 3G, WiFi, CDN, IPTV, DSL,Backbone networks and services such as Baidu.com. 50+ Papers, 18 issued US Patents
• Supervised interns from CMU, Cornell, Princeton, UCLA, GaTech, Michigan, Northwestern etc. Now @ Google, MSR, IBM, Bell Labs, Ohio State, Northeastern, HKUST etc. Currently supervising a dozen of Ph.D. students at THU
• Current research focus: Network Management for Mobile Internet
• Research Methodology: solve Real problem, build Real systems, process Real data with Big Data analytics.
4
Course Requirements• Goal: learn how to solve real network management
problems
http://netman.cs.tsinghua.edu.cn/~peidan/ANM
• Encourage interaction and discussion --- stop me and ask questions at any time!
• Most Importantly, enjoy the class!
5
Course Info• Time: Wednesday 9:50am-12:15pm
– Session 1: Network Management Fundamentals• Network Basics, Tools, Techniques
– Sessions 2 &3: Case Studies based on Conference Papers
• Office Hours: – Advise on your course projects. – By appointments, preferably during Thursday 2pm-
4:30pm.
6
What do you expect to learn (I): Networks & Services
• Targeted Services: Web-based Services such as search engine, online shopping and social networking; Video Streaming Services.
• Targeted Networks: Enterprise WiFi Network, Residential WiFi &Broadband Networks, Cellular Networks, and Data Center Networks.
• Targeted Devices: Smart Phones.
Enterprise WiFi Network s
Backbone
Data center
APController
7
What’s “Network Management”?
• A network management system utilizes real data to impact how network runs, but does not need to change existing software on vendor/user devices.
• It’s BIG DATA analytics with feedback loop!
8
Goals of Network & Service Management
– what happened in the past can be reconstructed automatically and accurately
– What’s going on now can be detected/inferred accurately to trigger automatic mitigation
– what will happen in the future can be predicted with some confidence
9
What do you expect to learn (II)?Network Management Areas
• Measurement
• Anomaly detection
• Diagnosis
• Mitigation.
10
What do you expect to learn (III)?Techniques
• Time series analysis
• Association rule mining
• Machine learning
Page 11
Enterprise WiFi Network s
Backbone
Data center
APController
This year’s focus:Managing The Performance of Mobile Internet
12
HTTP and Video Streaming Services
• Search Response Time [SIGCOMM 2013]
• Web QoE modeling [Mobicom 2014]
• Demystifying Web Page Load Time
• QoE model for video streaming [SIGCOMM 2013]
• Opprentince: automatic anomaly detection for Search Service
13
Home Networking
• Broadband Performance Measurement [Sigcomm2011]
• Performance Improvement [IMC2013]• What’s going on? [IMC2013]• Residential WiFi Interference [RxIP]• Residential WiFi Measurement [WiSe]• WiFi Union
14
Enterprise WiFi
• Campus WiFi measurement[Dartmouth]
• Campus WiFi troubleshooting [UCSD]• Inferring interference in Enterprise
WIFI [PIE]• Mobility Support in WiFi [SensorHints]
Smart Phones• Radio Resource on Smart Phone
– Radio Resource Allocation [ARO]– Periodical Transfers– Screen-Off Traffic
• Measurement from your smart phones– Crowdsourcing & Performance Monitoring [3GTest]– Realtime performance forecasting [PROTEUS]
• Caching on smart phone? [Ideal vs Reality]
• WiFi Tethering[DozyAP]
4-15
16
Cellular
• Carrier Measurement Methodologies– Throughput [Throughput Index]– Delay [TTFB]
• Content Caching– What to cache? [OTT video]– What caching technology? [MonBot]
• Accounting– Used Traffic Volume = Charged Traffic
Volume?
17
Data Center
• Failures– Measurement [SIGCOMM 2011]– Analyzing trouble tickets [NetSieve, NSDI
2013]– Service Impact [SoCC 2013]
• Traffic & Performance– Nature of Traffic [IMC 2009]– Traffic Characteristics [IMC 2010]
18
Course Projects• A project of team of one or two students.
• The course will provide a few suggested project topics, but the project team can choose their own.
• The instructor will interact with each project team regularly to help them pick project topic, and finish the project.
• Each Project will have one mentor (one of my Ph.D. students).
Page 19
Enterprise WiFi Network s
Backbone
Data center
APController
A glance at my research projects:Managing The Performance of Mobile Internet
How Much Are Your Neighbors Interfering with Your WiFi Web Experience?
20
Web Search
VoIPIM(WeChat,InstaMag) Social Media: Twitter,Weibo
500msdelay causes 1.2% decrease in Bing revenue[souders 2009]
Every Millisecond Counts!
400msdelay causes 0.74% decrease in Google searches[Brutlag 2009]
100msdelay causes 1% decrease in Amazon revenue[Linden 2013]
21
Wi-Fi Hop Latency
22
Wi-Fi Hop Latency
23
Anomaly Detection
24
25
26
Top-down Project Approaches:
• Top-down: start with the problem YOU want to solve.
• Your ideas? – Start with your pains when you using mobile
Internet, e.g., in Tsinghua Campus.
27
Suggested Project Ideas (I)• WiFi Security Scan: builds an App which scans the currently
connected WiFi network to see whether it is secure (e-payment, login, browsing etc, and find the security holes of the AP (server probe the AP) )– Based on encryption protocols
– Can you login using the factory default account/password (e.g., admin/admin)?
– Can you login from the WAN interface? Other security holes
• Measurement results• Suggest security fixes
28
Suggested Project Ideas (II)
• We already have an OpenWrt-based tool which measures the wireless interference between neighboring APs.
• Build an APP which can visualize and analyze the results from the above tool.
29
Suggested Project Ideas (III)
• Build a real-time heat map for the Campus WiFi.
30
Suggested Project Ideas (IV)
• We have the rough location information for each device every 3 minutes.
• Build a mobility model based on Campus WiFi data– Movement frequency distribution– Inter-movement time distribution– Migration Pattern
31
Suggested Project Ideas (V)
• Anomaly Detection Based on WiFi Usage data
• Detecting abnormal behavior by applying some existing anomaly detection algorithms (with existing implementation code)
32
Suggested Project Ideas (VI)
Build an idiot-proof Android App which can administrate an OpenWRT based router.
Some sample code:https://forum.openwrt.org/viewtopic.php?id=35574