Transcript A Survey of Active Network Research

A Survey of Active Network
Research
By:Tennehouse,Smith,Sincoskie,Wettherall,Minden
Presented By:Prashant, Ravikiran, Ashutosh
Objective
•Introduction to Active Networks
- Motivation
- Aims
- Impact
•How to Implement Active Network
- Discrete Approach
- Integrated Approach
• Snapshot of current research
Active Network
Active network is an approach to network
architecture in which the nodes of the network
perform customized computations on the message
flowing through them.
This concept permit applications to inject programs
into the nodes of network.
Example :
1) One can load “trace” program to routers and
program will be executed then and there when his
packets are processed at that router(active).
Origin of Active Networks
It is the result of discussion within Defense Advance
Research Projects Agency(DARPA) research
community in 1994-95.
Motivation for Active Network
• Difficulty of Integrating new technologies and
Standards.
• Poor performance because of redundant
operations at several protocol layers.
• Problems in accommodating new services.
Existing Applications :
Many application use a sort of Active Network
approach to handle current network related problems.
e.g.
•Firewall
•Web Proxies
•Nomadic Routers
•Application Services
Technology Advancement:
Recent advances in programming languages, compiler,
and operating systems provides the safe and efficient
execution of mobile programs.
Aim of Active Network:
• Programmable nodes.
•Standard execution environment
• No Standard functions
Impact of Active Networking
Accelerating infrastructure innovation
A new technology from prototype
demonstration to large scale deployment takes
about 10 years.Current backlog are:
. RSVP
. IPv6
. Multicast services
But Active networks eliminate the need for
formal standardization.
Enabling new technologies
• Merging and Distribution of information
N sources sending signals to M destinations.
End nodes does all of its mixing,work
With Active networks, work
 N
 MN
•Stock Quotes:
At present generally web caches do not cache the
stock quotes because of its dynamic nature and
even if it caches, the granularity of the object(entire
web page) is inappropriate.
Active networks can cache quotes at network nodes
using a per-stock name granularity.So, all client
request can be fulfilled with specified degree of
currency
• Online Auctions:
Servers responds to the current price requests but
due to delay in the network the information carried
by the packet may become outdated when it reach
the client. Then, the auction server will receive bids
that are too low and need to be rejected and this
imposes overhead on the heavily busy server.
With active networks these bids can be filtered out
before they reach the server. So the response time of
bid rejection to the client and server processing
overhead is reduced.
User Aware Network Protection
Active network may admit the design of integrated
mechanism to govern network resources and
information flowing through them. So, program in
Network security policy for the network on a peruser or per-use basis is allowed.
Active Network Management
Active technologies could be used to implement
sophisticated approaches to do network monitoring
and uninteresting event filtering.
Rethinking Performance
•Despite increasing the amount of processing at
nodes, Active network will lead to improved
performance.
i.e. Reduced throughput and increasing latency appear
to degrade the performance, they may actually
improve performance because of reduce demand of
bandwidth at endpoints, network congestion.
•Network performance is not always positively related
to the Application performance.
Caching in the memory as in the stock quote
example can reduce latency of data access when
the server is busy.
When network nodes in the auction application
reject low bids, they inform the “losing” end nodes
than could the overloaded (and farther away)
server.
Performance also depends on the location where the
active node is deployed.
•In the sensor fusion example, the greatest decrease
in bandwidth utilization occurs when the splitting of
multicast streams is performed as late as possible
and mixing as early as possible
•In stock quote example, it is important to place the
caches where they will serve the large number of
client request.
•In online auction,filter should be far enough from
the server to turn back low bids asap, but close
enough to the server to get proper up-to date price.
Approaches to implement the Active
Network
Programmable Switches- A discrete approach:
User would first inject their custom processing
routines into the required routers. Then they would
send there packets through such “programmable”
nodes. When the packet arrives at the node its header
is examined and the appropriate program is executed.
Capsules-An Integrated Approach:
In this approach ‘program’ is integrated in the packet
along with the data. When these capsule arrive at the
active nodes then it interprets the program and sends
the embedded data depending on its interpretation.
This is same to the Postscript code, where actual data
is embedded in program fragments that the printer
understands.
Active IP:
Active option is in the payload of the packet, the
legacy router can route the packet transparently to
active node where the ACTIVE Option code will be
evaluated and executed.
Common Programming model
Program Encoding:Our objectives of program
encoding are that they support
•Mobility- transfer programs and to execute them
on different platforms
•Safety-restrict resources that program can access
•Efficiency-enabling above without compromising
network performance.
Mobility can be achieved at different level of
program representation.
•Source level- Use of scripting language say TCL.
•Intermediate level- Use of Byte code virtual
instruction say in JAVA.
•Object level- Use binary formats as in Omniware.
Common Primitives:
The services built into each node might include
several categories of operations: primitives that
provide access to the node’s environment (e.g node
address, time of day and so on).
Node resources and their allocation:
There must be a common model of node resources
like bandwidth, processing capacity and the means by
which policies governing their allocation are
communicated.Safe resource allocation is a
considerable research area.
Current research
The goal of research is to develop:
•Languages and compiler for “active” software.
•Platforms for deploying Active Network nodes.
•Architecture for capsule and programmable
switches
•Safe resource allocation schemes.
•Massachusetts Institute of Technology-This team is
prototyping an architecture based on the capsule
approach and studying “active storage”, NACK fusion
and filtering.Demand loading is also a concern.
•University of Pennsylvania-The “Switch Ware”
project proposes a switch, which has a programmable
element performing switching function.
•Bell Communication Research- Several aspects of
the Penn design will be studied jointly with Bellcore
using different infrastructure OPCV2.
•Columbia University-Netscript architecture which
uses encapsulation approach along with scripting
language is developed.
•Carnegie Mellon University-It is developing
resource management mechanism in support of
“application aware” networks.
•Other Sites-University of Arizona, Georgia institute
of technology, University of Kansas, BBN,
University of Cincinnati.