Design and Implementation of HTTP-Gnutella Gateway

1
Design and Implementation of HTTP-Gnutella
Gateway

• Baoning Wu (baw4)
• Wei Zhang (wez5)
• CSE Department
• Lehigh University

2
Motivation

• Peer-to-peer networking is a hot topic.
• Can P2P nodes search and get files from Web
  sites?
• Can one P2P network search and get files from
  other P2P networks?
• In our project, we have built a special gateway
  between Gnutella and Web sites.

3
Related Work

• David McNab has launched Freenet search engine.
• Asiayeah is a Gnutella search engine.
• Filedonkey.com is an Edonkey search engine.
• Kalepa Networks , Inc is doing work about
  connecting different P2P systems.
• Our work is kind of reverse to all above works.

4
Mechanism of Gnutella Searching

• Node A sends a query to its neighbor B
• Node B boardcasts the query to its neighors C, D
  
• Node C has the objects node A needs and then
  returns a query hit message to node B
• Node B forwards the query hit message by
  consulting the local states.

5
Architecture of HTTP-Gnutella Gateway
6
Mechanism of the gateway

1. Node A broadcasts a query message directly or
   indirectly to the HTTP-Gnutella gateway
2. The HTTP-Gnutella gateway forwards the translated
   query message to search engine
3. The search engine returns a bunch of query
   results to the gateway
4. The gateway translates the results into Gnutella
   formats and then forwards them to node A
5. If node A initializes a download requests to the
   gateway, the gateway will translate the Gnutella
   request into a well-formatted HTTP request to the
   Web server
6. The gateway fetches the data from the Web server
   
7. The gateway forwards the data from the Web server
   to node A.

7
Handle Query Messages

• We still use the original Gnutella mechanism to
  judge whether to forward the message or not.
• The gateway captures all of queries with hops lt
  5 and sends them to search engine.

8
Search Engine API

• Google search engine API has a limit of up to
  1,000 requests per day.
• Search engine API consists of three main
  functions
• Query conversion
• Extraction of URLs
• Measurement of content size

9
Generate Query Hit Messages

• Two considerations
• Let Gnutella nodes contact Web servers directly
• Let the gateway work as a proxy
• The gateway fills its own IP address and a
  specific port number (currently 9999) in the
  query hit messages.
• File names are URLs of Web objects.

10
Downloading Service

• Translate Gnutella download request into a
  well-formatted HTTP request. e.g.
• GET /get/1234/http//www.foo.com/
  foo.mp3 HTTP/1.1
• User-Agent Gnutella
• Host 123.123.123.1236346
• gt
• GET http//www.foo.com/foo.mp3
  HTTP/1.1
• User-Agent Gnutella
• Host www.foo.com
• It should handle Gnutella handshakes properly.
• It also records the bytes transferred.

11
Problems Solutions

• Irregular handshakes
• We handle all possibilites
• File size
• We use HTTP HEAD request to get file size
• Broken Pipe signal
• We use forked process

12
Experiment Results

• Outline
• Basic verification and validation
• Log file format
• Results 1 to 4

13
Basic Verification Validation

• Run our special gateway on machine 1 and run a
  normal gtk-gnutalla client on machine 2. After
  machine 2 connects to machine 1, we use machine 2
  to send query messages and downloading request to
  machine 1.
• For downloaded files from machine 1, we use wget
  to get the same file from web server directly and
  use diff to test if they are identical.

14
Log File Format

• Log 1
• Time stamp, MUID, IP address, Type, Query
• Log 2
• Time stamp, IP address, URL, Size, Code, Success

15
Results 1

• No. of Query messages 319,245
• No. of Query Hit messages 930,860
• No. of served requests 113,391
• Average Response Time 16.33 seconds

16
Result 2
17
Result 3

• No. of Downloading requests 952
• No. of Different IP addresses 67
• No. of served Requests 945
• No. of sucessfully served requests 740
• Total size transfered 244,227,881 bytes
• Average response time 3.15 seconds
• Average total download time 15.92 seconds

18
Result 4
19
Future Work

• Support a variety of file types and measure their
  popularity
• Build a gateway to connect different P2P systems
• Deployment of such gateways

20
Conclusion

• An HTTP-Gnutella gateway was built and worked for
  the Gnutella users.
• Only 5 days, the gateway transferred about 244MB
  data from the Web sites to the Gnutella nodes.
• The systems achieved all goals of our design.

21

• Question?