CCN Reflects the Way How People Would Like to Use Internet Currently

CCN is a new internet architecture promoted by Van Jacobson[1]. After reading those materials [2,3,4] and comparing it with our current architecture, I think the most important difference is: under CCN architecture we focus on data without corncern with wherever it is; whereas under current architecture we first look for where the data is then touch the data. We have some techniques which can hide the extra step of looking for to enable users to focus on data without worry about location, for example overlay network.

Is one extra step big deal? The answer is it depends. If you only very rarely come across a simple scenario which need the step, It doesn't matter. However when it comes to internet, which is overwhelingly huge, extremely complicated and plays significantly roles in people's life, it realy matters. The reason is, for internet --- such a giant complex infrastructure, one more step means huge overhead and inconvenience sometimes even danger.

In general, I think it is the trend that we eventually will switch to this new architecture. Although current architecture may be still useful for some specific application, however, for the most important application --- information delivery --- we will use the new architecture. About the time frame, It is difficult to predict because it involves many factors. I personally belive in that once all technique challenges are resolved and the cost of switch is small enough comparing to its inconvinence, it will happen.

A Little History Relfecting the Trend

In the early year when internet just began to expand, the original goal was to connect together the original ARPANET with the ARPA packet radio network, in order to give users access to the large sevice machines on the ARPANET [5]. In other words, out goal is connect to the machine and work on it; data is not under consideration. The TCP/IP protocol works so well for that goal.

Later, email appeared [6] which means we were turnning to data; but we still felt comfortable with host-to-host style.

Later, WWW appeared[6]. We become further caring more about data. If you compare the sturcture of WWW URI with naming system of CCN, you can see they are almost in the same sturcture except there is a host part resolved by DNS in URI. However as a user, we don't care how you resolve the host part under the hood; we only care about the refered object. In some sence, WWW already has been telling us "What we care about is data".

As time flows, people pursuing other way to share data on internet developed P2P such as Freenet, Gnutella, Napster, and DHT(Distributed Hash Table) infrastructures.

One common characteristic of P2P and DHT is that they are using overlay network to hide the host-to-host details to enable users to focus on data. Nowadays the Internet is the basis for more overlaid networks that can be constructed in order to permit routing of messages to destinations not specified by an IP address [7].

Another sign that people are activly looking for a way working around host-to-host style is the emergence of Content Addressable Network, Chord, Pastry, and Tapestry Why did we bother to spend efforts on working around host-to-host style? I guess the reason probably is we have felt that the inconvenience of host-to-host has grown to be a problem.

Other problems coming with current architecture also indicate the TCP/IP-based architecture doesn't fit well in today's application. For examples, we have to spend lots of efforts to enable mobility; to develop CDN to guarrentee avalibiltiy; to consturct PKI system for security.

Today's goal is much different from the one at the beginning and the TCP/IP seems no longer match demands well.

How current network architecture works

Actual structures of internet are much complicated, here is a brief model.

First, local computational devices such as desktops, laptops, cell phones are phsically connected either with wire or wirelessly through switches, which form a local network; then these local networks are phsically connected across the world through gateways.

Second, to locate each device, we need an id for each of them, which is IP address. To enable them to communicate with each other, we need a "language" which is packets.

Third, we need a routing method to find the way between two devices on internet enabling them to touch each other indirectly or directly. This is a set of algorithms combining with IP address and packets form one of the most important protocol of current architecture --- IP.

Fourth, another most important protocol, TCP, protocol provides a higher-level interface for upper applications.

Now we are online, you can talk to anyone as long as you know its IP address or its DNS name which can give you the IP address under help of DNS server.

Some other protocols or techniques are there making above procedure faster or enabling it to go well in some extent without knowing IP address. For examples, CDN and P2P. Some other techniques or protocols make the internet more corlorful such as VOIP, Vedio stream. However everything is built on the TCP/IP or similar variants.

Gradually I cannot live without internet because I consult everything with some devices on internet. For any problems, always some devices has answer there on internet. How come if I don't know either the devices' name or its address? Search engines will help me almost completely although not exactly completely; since even search engines cannot extract all information while traversing each IP address. What if all search engine fail some day?

Problems with toady's architecture

Although today's architecuture works well, there are still many problems. For example, it cannot provide better abstraction for many applications such as multicast; the BGP is slow to converge and react to failures; the addressing method become a problem when hosts move.

The strategy of overlay network betters the situations in some extent. For example, it provide better abstraction by hiding IP layer; it is more reliable by keep multiple path and detect the traffic.

However there are still some problems for overlay network.For example the IP layer architecture limit the best performance of overlay network; additional layer make it more likely for some attacks such as man-in-the-middle attacks.

How CCN works

Here is a very brief model. I may misunderstand something. I try to describe it as far as I can.

First, we still need a physical connection which is similar to current TCP/IP architecture.

Second, We still need to identify what were accessing, but with CCN what we identify is data not a device; because we don't care on which devices the data is. Currently the id will be a heirachical structures of string for people with a binary encoding for machine. We still need a "language" which is similar to the packet in IP protocol except some changes due to the new naming and security methodology. The new packet contain signature, which in some extent takes on the role of PKI.

Third, for routing the path between source and destination, we still need a set of algorithms. The algorithms themselves probably similar to IP protocol, but the data structures storing the information for these algorithms dramatically change. The changes mainly are due to the new naming methodology.

Other important changes. On one hand there is a security layer which makes the CCN protocol inherently safer than TCP/IP architecture; on the other hand there is a strategy layer with tasks similar to TCP but much more flexible. By safer, I mean the security will be implemented in content not hosts which reduces the trust we must place in network intermediaries , opening the network to wide participations [2]. By more flexible, I mean you can try various methods to support reliability and flow control, mobility and connectivity and so on.

In addition, I am also confused about what if I don't know exact name of data I want. I just guess there may be a matching mechnism built in the protocol, which would eliminate our partial need for search engine.

Concerns about CCN

Although I personally belive in that it is trend to switch to CCN, there are still many challenges to solve. For examples, how to routing with high effeciency, how to cache contents to reduce congestions, how to further enhance security, how to further improve scalability. All of these questions need deep research.

Impacts of CCN

Following are impacts brought by CCN, I may be wrong about some opinions. Just try my best to understand them.

• Impacts on WWW

  I think the CCN is very similar to WWW on naming and retriving informaion. Therefore I don't think current architecture of WWW need significant changes. The main change probably is that we can directly resolve the URI without visiting DNS.

• Impacts on Overlay Network

  The CCN inherently possesses characteristics of overlay network while securer. Once CCN is adopted, many applications based on overlay network can easily be adapted to CCN.

• Impacts on Content Delivery Network

  The cache in CCN has played similar role to CDN. Maybe CCN can include CDN as its component and distributed it across the network.

• Impacts on Search Engine

  CCN should be able to take on simple searching tasks without the need to crawl across the whole internet.

Design Lesson from CCN

• Naming

  As our comp 150IDS told us, naming is so important which determine how you design and implement the whole system. We can consider URI and hierachical string used in CCN as naming. It possesses many advantages: 1> it facilitates lookup costing O(log n);
  2> it dicentralizes the right of naming; 3> it can be extensible...

• Module and Layer

  To design a complicated system, we must modularize the system and exploit layering to make the design more clear. By modulizing and layering, we focus on one important problem each time, which make implementation and maintenance more effecient. And also they enable us to modify or extend the system without affecting other parts.We can see the good examples from both CCN and TCP/IP architecture layers

• Performance

  Performance is important to every system. Beyond promoting performance by using advanced technology, great design is another valuable approach to achieve high performance. In CCN and current architecture, the cache is here and there. However we must take care about how to achieve coherence

• Scalability

  The Metcalfe's law tell us that the larger internet will be much more valuable. However, how to guarrentee the network will work as well as it grows dramatically is a big problem---scalability. Scalability is especially important to network architecture, which inherently has high requirement on this. In network architecture design, the designer are doing their best to exploit parallel methods and hierachical structure to reach high scalability. However the Amdahl's law always constrains the parallel extent.

• Security

  The network architecture didn't consider much security when they were built decades ago, since at that time the network was not very popular. Today when you are designing any system, you have to prepare for attack at the very beginning. The CCN seems more effecient on security than current architecture.

• Trade-off

  Any designe has traed-off, just as CCN choses content centric way while current architecture chosen host centric way.

Wrap ups

Nowday, we more and more prefer a content centric network. However designing a scalable, secure, giant system with high performance raise huge challenges. The new architecuture will take long time to evolve and become perfect. Finally people will get the network work better.