File Distribution Performance

Report
Authors: Haowei Yuan and Patrick Crowley
Publisher: 2013 Proceedings IEEE INFOCOM
Presenter: Chia-Yi Chu
Date: 2013/08/14
1




Introduction
Experimental Setup
File Distribution Performance
Improving CCNx Performance
2

Name-centric network architectures
◦ Data requests need to have unique names
◦ In-network storage elements that can cache the data and
respond to matching requests.

Named-Data Networking (NDN)
◦ Interest packets
 containing the name of the requested content
◦ Data packets
 containing both the name and its associated data
◦ NDN routers cache Data packets
 Entries in a cache indexed by their names.
3

HTTP infrastructure
◦ URLs are the names that matter most in today’s Internet.
◦ The requested URL in the HTTP header is the content name.


Including both web servers and caching proxies, can be
viewed as providing in-network storage for named
HTTP data.
Evaluate the effectiveness of NDN and HTTP as
content distribution systems over a range of
experimental scenarios.
4

Test bed
◦ Open Network Laboratory (ONL)
◦ 48 single-core machines
 AMD 2.0GHz Operon Processor, with 512MB memory and 1Gbps
network interface
◦ Connected via virtual switches Network Processor-based
Routers (NPRs)
5

CCNx Software Tools
◦ ccnx-0.4.0, release on Sep. 15, 2011.
◦ ccnd daemon
 Configured with default
 underlying transportation protocol is TCP
◦ Built-in ccncatchunks2
 Generate a sequence of Interest packets
◦ ccnfileserver
 Generate Data packets with content fetched from files on server
6

HTTP and Web-Caching Software Tools
◦ Lighttpd-1.4.28
◦ Squid-3.41.11
 Both using default configurations
◦ wget
 For downloading files
7

The metric
◦ Download Time (DT)
◦ the time from when a client application sends a request for a
file until the file is downloaded completely.
8

Experimental Configuration
◦ 40 client hosts, 1 server, and 2 levels of intermediate nodes
◦ 8 clients form a cluster, and shared a common second level
intermediate node
◦ Connected via 1Gbps links
◦ 100MB file is stored in server, clients try to fetch file
simultaneously
9
10

CCNx vs. Lighttpd
◦
◦
◦
◦
downloading 100MB file
without a caching proxy
Start with 1 client in each cluster
Active 1 clients each round until all clients are active
11
12

CCNx vs. Squid
◦ Single level case
 all the clients connect to the server through the top level CCNx
router or Squid proxy
◦ Two level case
 clients are connected via a second level cache
13
14

Lossy Network Condition
◦ Emulate a lossy link
 Rand drop plugin, which probabilistically selects and drops
packets on the NPRs.
◦ Emulate delay
 Delay plugin to an NPR connected with the link.
◦ 1 MB file
15
16
17
18


CCNx employs an XML encoding scheme to encode
packets to wire format.
The original CCNx implementation
◦ stores content with their names encoded in the Content Store
(CS)
◦ when the CS is queried, several content names might need to
be decoded

A simple change
◦ decoded content names are stored in the CS.
19
20

similar documents