1mb copy of newdoc

MS-DOS

• In the 1980s or early 1990s, the operating system that shipped with most PCs was a version of the Disk Operating System (DOS) created by Microsoft: MS-DOS. e

Peer-to-Peer

15-441

Outline

• p2p file sharing techniqueso Downloading: Whole-file vs. chunkso Searching

Centralized index (Napster, etc.) Flooding (Gnutella, etc.) Smarter flooding (KaZaA, …) Routing (Freenet, etc.)

• Uses of p2p - what works well, what doesn’t?o servers vs. arbitrary nodeso Hard state (backups!) vs soft-state (caches)

• Challengeso Fairness, freeloading, security, …

MS-DOS

• In the 1980s or early 1990s, the operating system that shipped with most PCs was a version of the Disk Operating System (DOS) created by Microsoft: MS-DOS. e

MS-DOS

• MS-DOS is a disk operating system for IBM PC–compatible computers.

• In its day, it was easily the most popular operating system in the world. It also is the basis from which Windows 9x derives its under-lying organization.

MS-DOS

• As with any other operating system, its function is to oversee the operation of the system by providing support for executing programs, controlling I/O devices, handling errors, and providing the user interface.

MS-DOS

• MS-DOS is a disk-based, single-user, single-task operating system. These qualities make it one of the easiest disk operating systems to understand.

MS-DOS

• The main portions of MS-DOS are the IO.SYS, MSDOS.SYS, and COMMAND.COM files.

• IO.SYS and MSDOS.SYS are special, hidden system files

MS-DOSThe IO.SYS file moves the system’s basic I/O functions into memory and then implements the MS-DOS default control programs, referred to as device drivers, for various hardware components.These include the following:• The boot disk drive• The console display and keyboard• The system’s time-of-day clock• The parallel and serial communications port

MS-DOS

The MSDOS.SYS file provides default support features for software applications.These features include the following:• Memory management• Character input and output• Real-time clock access• File and record management• Execution of other programs

MS-DOS

• The COMMAND.COM command interpreter accepts commands issued through the keyboard, or other input device, and carries them out according to the commands definition.

MS-DOS

• When DOS runs an application, COMMAND.COM finds the program, loads it into memory, and then gives it control of the system. When the program is shut down, it passes control back to the command interpreter.

MS-DOS

• The remainder of the operating system is comprised of utility programs to carry out DOS operations, such as formatting disks (FORMAT), printing files (PRINT), and copying files (XCOPY).

MS-DOS

• When MS-DOS was developed, it was designed to be a simple, stand-alone, operating system. To that end, it didn’t contain any network software, except SHARE.EXE.

• SHARE.EXE was designed as an add-on to popular networking software that allowed two users to edit the same file at the same time on a network.

MS-DOS

• Without SHARE.EXE , when a second user tried to open a file that the first user had opened, they got an error message. With SHARE.EXE installed, when the second user tried to open the file, they received a message saying that the file was being used by someone else and offering to provide a copy of the file.

MS-DOS• Another aspect of networking with MS-DOS is that

DOS can run client software for Novell and Microsoft networks. Most client software for DOS falls into the category of redirection software. This software redirects requests bound for local resources out to network resources.

• For example, with network client software installed, you could point a DOS drive letter to some disk space on the network. When you saved a file to that drive letter, you were really saving that file to a server. But, as far as DOS was concerned, it was accessing a local drive letter.

MS-DOS

MS-DOS Versions• Next we will look at the evolution of

the MS-DOS operating system and will examine the major changes in microcomputer architecture and standards that are reflected in each revision.

MS-DOS

MS-DOS 1• The original version of MS-DOS was, to

put it mildly, a “no-frills” operating system. It had no provisions for networking, did not include any sort of graphical shell program, and had limited ability to manage system resources. Approximately a year after the release of DOS 1.0, a revision—DOS 1.1—added support for double-sided 320KB floppy drives.

MS-DOS

MS-DOS 2• In early 1983, IBM introduced the IBM PC

XT. The XT featured a 10MB hard drive, a serial interface, and three additional expansion slots. It also had 128KB of RAM and a 360KB floppy drive (40KB more capacity than that of single-sided floppies on the previous PC) and could support a 10MB internal hard drive.

MS-DOS

MS-DOS 3• With DOS 3.0, released in summer 1984,

Microsoft continued to include additional DOS features and to support more powerful hardware. DOS 3.0 supported hard drives larger than 10MB, as well as enhanced graphics formats.

• Three revisions—3.1, 3.2, and 3.3—provided additional innovations.

MS-DOS

• Version 3.1 was notable because it featured the first DOS support for networking. The IBM PC Network was a simple local area network structure that was similar to today’s workgroup networks.

MS-DOS

• DOS 3.2 introduced the XCOPY command, enabling the user to identify more than one file at a time to be copied, and it made important modifications to other DOS commands. It was also the first version to support IBM’s Token Ring network topology and the first to allow for 720KB 3.5 Inch floppies.

MS-DOS

• Version3.3, introduced in 1987, offered additional enhancements to numerous existing commands and introduced support for 1.44MB floppy disks. Logical partition sizes could be up to 32MB, and a single machine could support both a primary and a secondary partition on each disk.

MS-DOS

MS-DOS 4• By 1988 it was apparent that the wave of the

future was the graphical interface, and DOS 4 provided users with the DOS Shell, a utility much like the Windows File Manager. Actually, DOS Shell was simply a scaled-down version of Windows that allowed users to manage files, run programs, and do routine maintenance, all from a single screen. The DOS Shell even supported a mouse.

MS-DOSMS-DOS 5• There were several important features

introduced in the 1991 release of DOS 5.0. First of all, the ability to load drivers into reserved (upper) memory was a relief to those people who were constantly running out of conventional memory. This feature allowed more complex DOS programs. Several software utilities made their debut. The most commonly used utility introduced at this time was EDIT.COM. Also added were QBASIC.EXE, DOSKEY, UNFORMAT, and UNDELETE.

MS-DOS

MS-DOS 6• Released in 1993 to excellent sales, DOS 6.0

offered a number of new commands and configurable options. Another enhancement in DOS 6.0 was EMM386.EXE, which allowed the system to pool extended and expanded memory. DOS 6.0 has subsequently been revised a number of times. As of this date, DOS 6.22 is the most current MS-DOS version available as a stand-alone operating system.

Wither p2p?

• Harness lots of spare capacityo 1 Big Fast Server: 1Gbit/s, $10k/month++o 2,000 cable modems: 1Gbit/s, $ ??o 1M end-hosts: Uh, wow.

• Build self-managing systems / Deal with huge scaleo Same techniques attractive for both companies /

servers / p2p E.g., Akamai’s 14,000 nodes Google’s 100,000+ nodes

P2p file-sharing

• Quickly grown in popularityo Dozens or hundreds of file sharing applicationso 35 million American adults use P2P networks --

29% of all Internet users in US!o Audio/Video transfer now dominates traffic on the

Internet

The p2p challenge

• C1: Search(human’s goals) -> fileo Given keywords / human description, find a

specific file

• C2: Fetch(file) -> bits

What’s out there?

DHTseDonkey2000New BT

KaZaA (bytes, not chunks)

BitTorrentChunkBased

FreenetGnutellaNapsterWholeFile

RouteSuper-node flood

FloodCentral

Searching

Internet

N1N2

N3

N6N5

N4

Publisher

Key=“title”Value=MP3 data… Client

Lookup(“title”)

?

Searching 2

• Needles vs. Haystackso Searching for top 40, or an obscure punk

track from 1981 that nobody’s heard of?• Search expressiveness

o Whole word? Regular expressions? File names? Attributes? Whole-text search? (e.g., p2p gnutella or p2p google?)

Framework

• Common Primitives:o Join: how to I begin participating?o Publish: how do I advertise my file?o Search: how to I find a file?o Fetch: how to I retrieve a file?

Next Topic...• Centralized Database

o Napster• Query Flooding

o Gnutella• Intelligent Query Flooding

o KaZaA• Swarming

o BitTorrent• Unstructured Overlay Routing

o Freenet• Structured Overlay Routing

o Distributed Hash Tables

Napster: History

• 1999: Sean Fanning launches Napster• Peaked at 1.5 million simultaneous

users• Jul 2001: Napster shuts down

Napster: Overiew

• Centralized Database:o Join: on startup, client contacts central

servero Publish: reports list of files to central

servero Search: query the server => return

someone that stores the requested fileo Fetch: get the file directly from peer

Napster: Publish

I have X, Y, and Z!

Publish

insert(X,123.2.21.23)...

123.2.21.23

Napster: Search

Where is file A?

Query Reply

search(A)-->123.2.0.18Fetch

123.2.0.18

Napster: Discussion

• Pros:o Simpleo Search scope is O(1)o Controllable (pro or con?)

• Cons:o Server maintains O(N) Stateo Server does all processingo Single point of failure

Query Flooding Overview

• Join: on startup, client contacts a few other nodes; these become its “neighbors”

• Publish: no need• Search: ask neighbors, who ask their

neighbors, and so on... when/if found, reply to sender.o TTL limits propagation

• Fetch: get the file directly from peer

I have file A.

I have file A.

Search in Query Flooding

Where is file A?

Query

Reply

Flooding Discussion

• Pros:o Fully de-centralizedo Search cost distributedo Processing @ each node permits powerful search

semantics• Cons:

o Search scope is O(N)o Search time is O(???)o Nodes leave often, network unstable

• TTL-limited search works well for haystacks.o For scalability, does NOT search every node. May

have to re-issue query later

Query Floding: Gnutella

• In 2000, J. Frankel and T. Pepper from Nullsoft released Gnutella

• Soon many other clients: Bearshare, Morpheus, LimeWire, etc.

• In 2001, many protocol enhancements including “ultrapeers”

Flooding with Supernodes

• “Smart” Query Flooding:o Join: on startup, client contacts a “supernode” ...

may at some point become one itselfo Publish: send list of files to supernodeo Search: send query to supernode, supernodes

flood query amongst themselves.o Fetch: get the file directly from peer(s); can fetch

simultaneously from multiple peers

Supernodes Network Design“Super Nodes”

Supernodes: File Insert

I have X!

Publish

insert(X,123.2.21.23)...

123.2.21.23

Supernodes: File Search

Where is file A?

Query

search(A)-->123.2.0.18

search(A)-->123.2.22.50

Replies

123.2.0.18

123.2.22.50

Supernodes: Fetching(And use of hashes…)

• More than one node may have requested file...• How to tell?

o Must be able to distinguish identical fileso Not necessarily same filenameo Same filename not necessarily same file...

• Use Hash of fileo KaZaA uses UUHash: fast, but not secureo Alternatives: MD5, SHA-1

• How to fetch?o Get bytes [0..1000] from A, [1001...2000] from Bo Alternative: Erasure Codes

Supernode Flooding Discussion

• Pros:o Tries to take into account node heterogeneity:

Bandwidth Host Computational Resources Host Availability (?)

o Rumored to take into account network localityo Scales better

• Cons:o Mechanisms easy to circumvento Still no real guarantees on search scope or search time

• Similar behavior to plain flooding, but better.

Stability and Superpeers

• Why superpeers?o Query consolidation

Many connected nodes may have only a few files Propagating a query to a sub-node would take more b/w

than answering it yourselfo Caching effect

Requires network stability• Superpeer selection is time-based

o How long you’ve been on is a good predictor of how long you’ll be around.

Superpeers: KaZaA

• In 2001, KaZaA created by Dutch company Kazaa BV

• Single network called FastTrack used by other clients as well: Morpheus, giFT, etc.

• Eventually protocol changed so other clients could no longer talk to it

• Most popular file sharing network in 2005 with >10 million users (number varies)

Searching & Fetching

• Query flooding finds:o An object

Filename? Hash?

o A host that serves that object• In QF systems, d/l from the host that

answered your query• Generally uses only one source…

Fetching in 2006

• When you have an object ID,• Get a list of peers serving that ID

o Easier than the keyword lookupo Queries are structured

• Download in parallel from multiple peers

• “Swarming”o Download from others downloading same

object at same time

Swarming: BitTorrent

• In 2002, B. Cohen debuted BitTorrent• Key Motivation:

o Popularity exhibits temporal locality (Flash Crowds)o E.g., Slashdot effect, CNN on 9/11, new movie/game release

• Focused on Efficient Fetching, not Searching:o Distribute the same file to all peerso Single publisher, multiple downloaders

• Has some “real” publishers:o Blizzard Entertainment using it to distribute the beta of their

new game

BitTorrent: Overview

• Swarming:o Join: contact centralized “tracker” server, get a list

of peers.o Publish: Run a tracker server.o Search: Out-of-band. E.g., use Google to find a

tracker for the file you want.o Fetch: Download chunks of the file from your

peers. Upload chunks you have to them.• Big differences from Napster:

o Chunk based downloading (sound familiar? :)o “few large files” focuso Anti-freeloading mechanisms

BitTorrent: Publish/JoinTracker

BitTorrent: Fetch

BitTorrent: Sharing Strategy

• Employ “Tit-for-tat” sharing strategyo A is downloading from some other people

A will let the fastest N of those download from himo Be optimistic: occasionally let freeloaders

download Otherwise no one would ever start! Also allows you to discover better peers to download from

when they reciprocateo Let N peop

• Goal: Pareto Efficiencyo Game Theory: “No change can make anyone

better off without making others worse off”o Does it get there? No, but it’s reasonable

BitTorrent: Summary

• Pros:o Works reasonably well in practiceo Gives peers incentive to share resources; avoids

freeloaders• Cons:

o Pareto Efficiency relative weak conditiono Central tracker server needed to bootstrap swarm o Tracker is a design choice, not a requirement.

Newer BT variants use a “distributed tracker” - a Distributed Hash Table

Next Topic...• Centralized Database (Searching)

o Napster• Query Flooding (Searching)

o Gnutella• Supernode Query Flooding (Searching)

o KaZaA• Swarming (Fetching)

o BitTorrent• Unstructured Overlay Routing (Both?)

o Freenet• Structured Overlay Routing (Both, but mostly

search)o Distributed Hash Tables (DHT)

Distributed Hash Tables• Academic answer to p2p• Goals

o Guatanteed lookup successo Provable bounds on search timeo Provable scalability

• Makes some things hardero Fuzzy queries / full-text search / etc.

• Read-write, not read-only• Hot Topic in networking since introduction in

~2000/2001

DHT: Overview• Abstraction: a distributed “hash-table” (DHT)

data structure:o put(id, item);o item = get(id);

• Implementation: nodes in system form a distributed data structureo Can be Ring, Tree, Hypercube, Skip List, Butterfly

Network, ...

DHT: Overview (2)

• Structured Overlay Routing:o Join: On startup, contact a “bootstrap” node and integrate

yourself into the distributed data structure; get a node ido Publish: Route publication for file id toward a close node id

along the data structureo Search: Route a query for file id toward a close node id. Data

structure guarantees that query will meet the publication.o Fetch: Two options:

Publication contains actual file => fetch from where query stops Publication says “I have file X” => query tells you 128.2.1.3 has

X, use IP routing to get X from 128.2.1.3

DHT: Example - Chord

• Associate to each node and file a unique id in an uni-dimensional space (a Ring)o E.g., pick from the range [0...2m]o Usually the hash of the file or IP address

• Properties:o Routing table size is O(log N) , where N is the total

number of nodeso Guarantees that a file is found in O(log N) hops

from MIT in 2001

DHT: Consistent Hashing

N32

N90

N105

K80

K20

K5

Circular ID space

Key 5Node 105

A key is stored at its successor: node with next higher ID

DHT: Chord Basic Lookup

N32

N90

N105

N60

N10N120

K80

“Where is key 80?”

“N90 has K80”

DHT: Chord “Finger Table”

N80

1/21/4

1/8

1/161/321/641/128

• Entry i in the finger table of node n is the first node that succeeds or equals n + 2i

• In other words, the ith finger points 1/2n-i way around the ring

DHT: Chord Join

• Assume an identifier space [0..8]• Node n1 joins

01

2

34

5

6

7i id+2i succ0 2 11 3 12 5 1

Succ. Table

DHT: Chord Join

• Node n2 joins0

1

2

34

5

6

7i id+2i succ0 2 21 3 12 5 1

Succ. Table

i id+2i succ0 3 11 4 12 6 1

Succ. Table

DHT: Chord Join

• Nodes n0, n6 join 0

1

2

34

5

6

7i id+2i succ0 2 21 3 62 5 6

Succ. Table

i id+2i succ0 3 61 4 62 6 6

Succ. Table

i id+2i succ0 1 11 2 22 4 0

Succ. Table

i id+2i succ0 7 01 0 02 2 2

Succ. Table

DHT: Chord Join

• Nodes: n1, n2, n0, n6

• Items: f7, f2

01

2

34

5

6

7 i id+2i succ0 2 21 3 62 5 6

Succ. Table

i id+2i succ0 3 61 4 62 6 6

Succ. Table

i id+2i succ0 1 11 2 22 4 0

Succ. Table

7

Items 1

Items

i id+2i succ0 7 01 0 02 2 2

Succ. Table

DHT: Chord Routing

• Upon receiving a query for item id, a node:

• Checks whether stores the item locally

• If not, forwards the query to the largest node in its successor table that does not exceed id

01

2

34

5

6

7 i id+2i succ0 2 21 3 62 5 6

Succ. Table

i id+2i succ0 3 61 4 62 6 6

Succ. Table

i id+2i succ0 1 11 2 22 4 0

Succ. Table

7

Items 1

Items

i id+2i succ0 7 01 0 02 2 2

Succ. Table

query(7)

DHT: Chord Summary

• Routing table size?o Log N fingers

• Routing time?o Each hop expects to 1/2 the distance to the

desired id => expect O(log N) hops.

DHT: Discussion

• Pros:o Guaranteed Lookupo O(log N) per node state and search scope

• Cons:o No one uses them? (only one file sharing

app)o Supporting non-exact match search is hard

When are p2p / DHTs useful?

• Caching and “soft-state” datao Works well! BitTorrent, KaZaA, etc., all use

peers as caches for hot data• Finding read-only data

o Limited flooding finds hayo DHTs find needles

• BUT

A Peer-to-peer Google?

• Complex intersection queries (“the” + “who”)o Billions of hits for each term alone

• Sophisticated rankingo Must compare many results before returning a

subset to user• Very, very hard for a DHT / p2p system

o Need high inter-node bandwidtho (This is exactly what Google does - massive

clusters)

Writable, persistent p2p

• Do you trust your data to 100,000 monkeys?• Node availability hurts

o Ex: Store 5 copies of data on different nodeso When someone goes away, you must replicate the

data they heldo Hard drives are *huge*, but cable modem upload

bandwidth is tiny - perhaps 10 Gbytes/dayo Takes many days to upload contents of 200GB

hard drive. Very expensive leave/replication situation!

P2P: Summary

• Many different styles; remember pros and cons of eacho centralized, flooding, swarming, unstructured and structured

routing• Lessons learned:

o Single points of failure are very bado Flooding messages to everyone is bado Underlying network topology is importanto Not all nodes are equalo Need incentives to discourage freeloadingo Privacy and security are importanto Structure can provide theoretical bounds and guarantees

Extra Slides

KaZaA: Usage Patterns• KaZaA is more than

one workload!o Many files < 10MB

(e.g., Audio Files)o Many files > 100MB

(e.g., Movies)

from Gummadi et al., SOSP 2003

KaZaA: Usage Patterns (2)• KaZaA is not Zipf!

o FileSharing: “Request-once”

o Web: “Request-repeatedly”


KaZaA: Usage Patterns (3)

• What we saw:o A few big files consume most of the bandwidtho Many files are fetched once per client but still very popular

• Solution?o Caching!


Freenet: History

• In 1999, I. Clarke started the Freenet project

• Basic Idea:o Employ Internet-like routing on the overlay

network to publish and locate files• Addition goals:

o Provide anonymity and securityo Make censorship difficult

Freenet: Overview

• Routed Queries:o Join: on startup, client contacts a few other

nodes it knows about; gets a unique node ido Publish: route file contents toward the file id. File

is stored at node with id closest to file ido Search: route query for file id toward the closest

node ido Fetch: when query reaches a node containing

file id, it returns the file to the sender

Freenet: Routing Tables• id – file identifier (e.g., hash of file)• next_hop – another node that stores the file id• file – file identified by id being stored on the local node• Forwarding of query for file id

o If file id stored locally, then stop Forward data back to upstream requestor

o If not, search for the “closest” id in the table, and forward the message to the corresponding next_hop

o If data is not found, failure is reported back Requestor then tries next closest match in routing

table

id next_hop file

…

…

Freenet: Routing

4 n1 f412 n2 f125 n3

9 n3 f9

3 n1 f314 n4 f145 n3

14 n5 f1413 n2 f133 n6

n1 n2

n3

n4

4 n1 f410 n5 f108 n6

n5

query(10)

1

2

3

4

4’

5

Freenet: Routing Properties

• “Close” file ids tend to be stored on the same nodeo Why? Publications of similar file ids route toward

the same place• Network tend to be a “small world”

o Small number of nodes have large number of neighbors (i.e., ~ “six-degrees of separation”)

• Consequence:o Most queries only traverse a small number of hops

to find the file

Freenet: Anonymity & Security

• Anonymityo Randomly modify source of packet as it traverses the networko Can use “mix-nets” or onion-routing

• Security & Censorship resistanceo No constraints on how to choose ids for files => easy to have

to files collide, creating “denial of service” (censorship)o Solution: have a id type that requires a private key signature

that is verified when updating the fileo Cache file on the reverse path of queries/publications =>

attempt to “replace” file with bogus data will just cause the file to be replicated more!

Freenet: Discussion

• Pros:o Intelligent routing makes queries relatively short o Search scope small (only nodes along search path

involved); no floodingo Anonymity properties may give you “plausible

deniability”• Cons:

o Still no provable guarantees!o Anonymity features make it hard to measure,

debug