Questioning and Routing Data-Centric Forays into Networking .


52 views
Uploaded on:
Category: Sales / Marketing
Description
Querying and Routing Data-Centric Forays into Networking. Joe Hellerstein UC Berkeley and Intel Research. Note. These slides were made on PowerPoint for Mac 2004 There are incompatibilities between the Mac and Windows versions of PowerPoint, particularly with regard to animations.
Transcripts
Slide 1

Questioning and Routing Data-Centric Forays into Networking Joe Hellerstein UC Berkeley and Intel Research

Slide 2

Note These slides were made on PowerPoint for Mac 2004 There are incongruencies between the Mac and Windows variants of PowerPoint, especially with respect to livelinesss. If you don\'t mind email the creator with inquiries.

Slide 3

Road Map Emerging cooperative energies in databases and systems administration Internet-Scale Querying: PIER and  Agenda, outline space Toward a Network Oracle (  ) The PIER Query Processor Design standards & challenges Overlay Networks: DHTs Query Processing on DHTs PIER in real life If time grants Routing with questions Related issues in Sensor Networks (TinyDB and BBQ)

Slide 4

Background: CS262 Experiment w/Eric Brewer Merge OS & DBMS graduate class, over a year Eric/Joe, point/counterpoint Some tie-ins were self-evident: memory mgmt, capacity, booking, simultaneousness Surprising: QP and systems go well next to each other Query processors are dataflow motors. So are switches (e.g. Kohler\'s CLICK toolbox). Versatile inquiry strategies look significantly more like systems administration thought E.g. "Swirl" tuple switches and TCP Congestion Control Use basic Control/Queuing to "learn"/influence unusual dataflows

Slide 5

Networking for DB Dummies (i.e. me) Core capacity of conventions: information xfer Data Manipulation support, checksum, encryption, xfer to/fr application space, presentation Transfer Control stream/blockage ctl, distinguishing xmission probs, acks, muxing, timestamps, confining Clark & Tennenhouse, "Compositional Considerations for a New Generation of Protocols", SIGCOMM \'90 Basic Internet supposition: "a system of obscure topology and with an obscure, mysterious and continually changing populace of contending discussions" (Van Jacobson)

Slide 6

Data Modeling! Inquiry Opt! Trade! C & T\'s Wacky Ideas Thesis: nets are great at xfer control, not all that great at information control Some C&T wacky thoughts for better information control Xfer semantic units, not parcels (ALF) Auto-change layers to smooth them (ILP) Minimize cross-layer requesting imperatives Control conveyance in parallel by means of bundle substance

Slide 7

Wacky Ideas in Query Processing What in the event that… We had unbounded information makers and buyers ("streams" … "consistent inquiries") We couldn\'t know our makers\' conduct or substance?? ("league" … "go betweens") We couldn\'t anticipate client conduct? ("CONTROL") We couldn\'t anticipate conduct of segments in the dataflow? ("web administrations") We had fractional disappointment as a given? (exchanges unrealistic?) Yes … organizing individuals have been here! Review Van Jacobson\'s quote

Slide 8

Convergence Data Models, Query Opt, DataScalability DATABASE RESEARCH P2P Queries Approximate/Interactive QP Adaptive Dataflow SensorNet Queries TinyDB BBQ PIER  Content-Based Routing Knowledge Plane Router Toolkits Wireless Meshes NETWORKING RESEARCH Adaptivity, Federated Control, NodeScalability

Slide 9

Road Map Emerging cooperative energies in databases and systems administration Internet-Scale Querying: PIER and  Agenda, outline space Toward a Network Oracle (  ) The PIER Query Processor Design standards & challenges Overlay Networks: DHTs Query Processing on DHTs PIER in real life If time licenses Routing with inquiries Related issues in Sensor Networks (TinyDB and BBQ)

Slide 10

PIER P 2P I nformation E xchange and R etrieval An Internet-Scale (shared) question motor

Slide 11

Internet Scale 1000\'s - Millions Single Site Clusters Distributed 10\'s – 100\'s Our story at VLDB: What is a Very Large Data Base? [HHLLSS VLDB 03] Challenge: How to run DB style questions at Internet Scale?! Challenge: How can DB usefulness change the Internet? Database Community Network Community

Slide 12

What are the Key Properties? Heaps of information that is: Naturally conveyed (where it\'s created) Centralized gathering undesirable Homogeneous in diagram Data is more valuable when seen all in all This is the outline space we have examined. Instead of … Enterprise Information Integration Semantic Web Challenges tilted all the more intensely toward frameworks/calculations rather than information semantics & cleaning

Slide 13

Who Needs Internet Scale Querying? Illustration 1: Filenames Simple pervasive patterns: Filenames, Sizes, ID3 labels Early P2P filesharing applications Napster, Gnutella, KaZaA, and so forth. Worked "in the carport" "Ordinary" non-master clients Not the best illustration Often used to break copyright Fairly inconsequential innovation But… Points to key social issues driving appropriation of decentralized frameworks Provide genuine workloads to approve more perplexing plans

Slide 14

Example 2: Network Traces Schemas are for the most part institutionalized: IP, SMTP, HTTP, SNMP log groups, firewall log positions, and so forth. Arrange managers are searching for examples inside their website AND with different destinations: DoS assaults cross managerial limits Tracking the study of disease transmission of infections/worms Timeliness is extremely useful Might shock you exactly how helpful this is: Network on PlanetLab (disseminated examine test bed) is for the most part loaded with individuals observing the system status

Slide 15

Road Map Emerging cooperative energies in databases and systems administration Internet-Scale Querying: PIER and  Agenda, outline space Toward a Network Oracle (  ) The PIER Query Processor Design standards & challenges Overlay Networks: DHTs Query Processing on DHTs PIER in real life If time licenses Routing with inquiries Related issues in Sensor Networks (TinyDB and BBQ)

Slide 16

 : Public Health for the Internet [HPPRSW 04] Thought test: A Network Oracle Queryable substance that thinks about all system state Network maps Link stacking Point-to-point latencies/transfer speed Event location (e.g. firewall occasions) Naming (DNS, ASs, and so forth.) End-framework setup Router design Data from later past up to close ongoing Available to all end-frameworks What may this empower?

Slide 17

Applications of a Network Oracle Performance blame finding Tracking system assaults Correlating firewall logs New steering conventions E.g. application particular course choice Adaptive disseminated applications "Web Screensavers" A la SETI@Home Serendipity!

Slide 18

Benefits? Short term: Connect net estimation and security scientists\' datasets. Empower disseminated questions for system portrayal, the study of disease transmission and alarms. E.g. best 10 IP address result from Barford et.al. Medium term: Provide an administration for overlay systems and planetary-scale versatile applications E.g. sustain interface estimation comes about into CDNs, server choice Long term: Change the Internet: conventions no longer expect obliviousness of system state. Push more knowledge into end frameworks. E.g. Have based source steering arrangements, blockage shirking (setting timeouts)

Slide 19

A Center for Disease Control? Who possesses the Center? What do they Control? This will be disliked, best case scenario Electronic protection for people The Internet as "an extensively surveilled police state"? Supplier disincentives Transparency = bolster cost, shame And difficult to convey Can screen the chokepoints (ISPs) But inside intranets?? E.g. Intel IT E.g. Berkeley dormitories E.g. Grassroots WiFi agglomerations?

Slide 20

Energizing the End-Users Endpoints are omnipresent Internet, intranet, hotspot Toward a uniform engineering End-clients will help Populist engage home clients is opportune Enterprise IT can manage endpoint programming Differentiating motivating forces for endpoint sellers The association: distributed innovation Harnessed to the great! Usability Built-in scaling Decentralization of trust and obligation

Slide 21

Road Map Emerging collaborations in databases and systems administration Internet-Scale Querying: PIER and  Agenda, outline space Toward a Network Oracle (  ) The PIER Query Processor Design standards & challenges Overlay Networks: DHTs Query Processing on DHTs PIER in real life If time grants Routing with inquiries Related issues in Sensor Networks (TinyDB and BBQ)

Slide 22

Coarse Architecture of PIER

Slide 23

Declarative Queries Query Plan Overlay Network Physical Network Q u e r y C o r e O p t i m i z e R e l a t i o n a l E x e c u t i o n C a t a l o g E n g i n e M a n a g e r P I E R

Slide 24

Road Map Emerging cooperative energies in databases and systems administration Internet-Scale Querying: PIER and  Agenda, plan space Toward a Network Oracle (  ) The PIER Query Processor Design standards & challenges Overlay Networks: DHTs Query Processing on DHTs PIER in real life If time licenses Routing with questions Related issues in Sensor Networks (TinyDB and BBQ)

Slide 25

Some Background on Overlay Networks [RH ITR 03] A P2P framework like PIER needs to: Track personalities & (IP) locations of companions at present online May be numerous! May have noteworthy Churn Best not to have n 2 ID references Route messages among associates If you don\'t track all companions all around, this is "multi-bounce" This is an overlay organize Peers are doing both naming and steering IP turns out to be "only" the low-level transport All the IP directing is misty

Slide 26

What is a DHT? Hash Table information structure that maps "keys" to "qualities" vital building obstruct in programming frameworks Distributed Hash Table (DHT) comparative, yet spread over the Internet Interface insert(key, esteem) lookup(key)

Slide 27

How? Each DHT hub underpins a solitary operation: Given key as info; course messages toward hub holding key

Slide 28

K V K V K V K V K V K V K V K V K V K V K V DHT in real life

Slide 29

K V K V K V K V K V K V K V K V K V K V K V DHT in real life

Slide 30

K V K V K V K V K V K V K V K V K V K V K V DHT in real life Operation: take key as information; course messages to hub holding key

Slide 31

K V K V K V K V K V K V K V K V K V K V K V DHT in real life: put() insert(K 1 ,V 1 ) Operation: take key

Recommended
View more...