wiki:OldGPGResearchChallenges

Version 1 (modified by hdempsey@bbn.com, 11 years ago) (diff)

--

*GENI PLANNING GROUP DOCUMENTS ARE NO LONGER CURRENT. See GpoDoc and the GENI project pages for up-to-date documents and specifications.*

Design Challenges

The research challenge at the center of GENI is to understand how to design and build a Future Internet that achieves its potential. The Internet has been a fantastic success, but in many ways it is not meeting the needs of its users. Today, it is not secure, hard to use, and unpredictable. Its technical design has created barriers rather than stimulants to key industrial investments. Tomorrow, it needs to support emerging computing technologies, new network technologies such as wireless, and emerging applications. Getting from where we are now to a better Internet for the future is a goal of critical national importance.

Security and Robustness

Perhaps the most compelling reason to rethink the Internet is to get a network with greatly improved security and robustness. The Internet of today has no overarching approach to dealing with security -- it has lots of mechanisms but no "security architecture" -- no set of rules for how these mechanisms should be combined to achieve overall good security. Security on the net today more resembles a growing mass of band-aids than a plan.

We take a broad definition of security and robustness. The traditional focus of the security research community has been on protection from unwanted disclosure and corruption of data. We propose to extend this to availability and resilience to attack and failure. Any Future Internet should attain the highest possible level of availability, so that it can be used for "mission-critical" activities, and it can serve the nation in times of crisis. We should do at least as well as the telephone system, and in fact better.

Many of the actual security problems that plague users today are not in the Internet itself, but in the personal computers that attach to the Internet. We cannot say we are going to address security and not deal with issues in the end-nodes as well as the network. This is a serious challenge, but it offers an opportunity to reach beyond the traditional network research community and engage groups that look at operating systems and distributed systems design.

Our most vexing security problems today are not just failures of technology, but result from the interaction between human behavior and technology. For example, if we demanded better identification of all Internet users, it might make tracking attacks and abuse easier, but loss of anonymity and constant surveillance might have a very chilling effect on many of the ways the Internet is used today. A serious redesign of Internet security must involve tech-savvy social scientists and humanists from the beginning, to understand the larger consequences of specific design decisions.

We identify the following specific design challenges in building a secure and robust network:

  • Any set of "well-behaved" hosts should be able to communicate among themselves as they desire, with high reliability and predictability, and malicious or corrupted nodes should not be able to disrupt this communication. Users should expect a level of availability that matches or exceeds the telephone system of today.
  • Security and robustness should be extended across layers. Because security and reliability to an end user depends on the robustness of both the network layer and the distributed applications.
  • There should be a reasoned balance between identity for accountability and deterrence and privacy and freedom from unjustified observation and tracking.

Support for New Network Technology

The current Internet is designed to take advantage of a wide range of underlying network technologies. It is worth remembering that the Internet is older than both local area networks and fiber optics, and had to integrate both those technologies. It has done so with great success. However, there are many new challenges on the horizon.

The current "new technology on the block" is wireless in all its forms, from WiFi today to Ultra-wideband and wireless sensor networks tomorrow. Wireless is perhaps one of the most transforming and empowering network technologies to come along, equal or greater in impact to the local-area network (LAN). For example, laptop sales exceeded those of desktop personal computers in 2003 and this trend towards compact and portable computing devices continues unabated. As of 2005, it is estimated that there are over 2 billion cell phones in use worldwide as compared with 500 million wired Internet terminals, and a significant fraction (~20%) of these phones now have data capabilities as 2.5G and 3G cellular services are deployed. In another 5 years, all cell phones will be full-fledged Internet devices implying inevitable changes both in applications and network infrastructure to support mobility, location-awareness and processing/bandwidth limitations associated with this class of end-user terminals. Clearly, we need to think now about how a Future Internet and new modes of wireless can best work with each other

The most obvious consequence of wireless is mobility. We see mobility today at the "edge" of the network, when we read our email on our PDAs. We have a weak form of mobility with our laptops today, where we connect sporadically to WiFi hot spots. But the Internet itself does not support these activities well, and indeed in most cases is oblivious to them. The default node on the Internet today is still the stationary PC on a desktop. We must rethink what support is needed for the mobile host.

Perhaps less obvious, but equally important, while wire-based technology such as Ethernet just keeps getting faster, some wireless technology (especially that which works in challenging situations) is slow and erratic. The power of "always connected" may be accompanied by the limitation of unpredictable performance. We must think through how applications are designed to work in this context, and how a Future Internet can best support this wireless experience.

Similarly, because the devices connected to wireless networks must be power aware, and dynamic spectrum give wireless devices an extra degree of freedom in how they utilize the communication medium, fundamental changes are needed in how we think about the network. The Future Internet must support adaptive and efficient resource usage, for example, by treating links not just as a rigid "input", but as a flexible "parameter" that can be tailored to meet the needs of the user.

Mobility increases the need to deal with issues of dynamic resource location and binding, and the linking of physical and cyber-location. In general, the network must support location awareness; the ability to exploit location information to provide services should be incorporated throughout the network architecture.

Finally, we need to understand the design principles for wireless networks in an Internet context. Like the Internet, the most popular wireless protocols today are insecure, fragile, hard to configure, and poorly adapted to support demanding applications. As just one example, the security of the popular 802.11 WiFi standard has been shown to be vulnerable to systematic attack. We need to build realistic, live prototypes to point the way to addressing these fundamental problems with today's wireless technologies.

We identify the following specific design challenges in supporting wireless technology:

  • A Future Internet must support node mobility as a first-level objective. Nodes must be able to change their attachment point to the Internet.
  • A Future Internet must provide adequate means for an application to discover characteristics of varying wireless links and adapt to them.
  • A Future Internet (or a service running on that Internet) must facilitate the process by which nodes that are in physical proximity discover each other.
  • A Future Internet must develop wireless technologies that work well in an Internet context, with robust security, resource control, and interaction with the wired world.

A second technology revolution is taking place in the underlying optical transport, where the optics research community is about to undergo a dramatic shift, roughly equivalent to that of the electronics community in the early 1960s. Optical communications researchers are discovering how to use new technologies like optical switches and logic elements to deliver much higher performance at lower power than purely electronics solutions.

In particular, the advent of large-scale electronic integration that took the world by storm and led to the PC and wireless foreshadows a revolution that is about to take place with optics (photonics). The photonic integrated circuit (PIC) is allowing ever-increasing complexity in optical circuits and functions to be placed on a single chip alongside electronic circuits, to enable networking and communications paradigms not possible with electronics alone. As PIC technology matures, it will enable higher capacity networks that are reconfigurable, more flexible and have much higher capacity at much lower cost. This may involve moving from ring to mesh networks, from fixed wavelength allocations to tunable transmitters and receivers, from networks without optical buffering to ones with intelligent control planes and sufficient optical buffering, and from networks that treat fiber bandwidth as fixed circuits to networks that allow the fiber bandwidth to be dynamically accessed and utilized.

We identify the following specific design challenges in exploiting emerging optical capabilities:

  • A Future Internet must be designed to enable users to leverage these new capabilities of the underlying optical transport, including better reliability through cross-layer diagnostics, better predictability at lower cost through cross-layer traffic engineering, and much higher performance to the desktop.
  • A Future Internet must allow for dynamically reconfigurable optical nodes that enable the electronics layer to dynamically access the full fiber bandwidth.
  • A Future Internet must include control and management software that allow a network of dynamically reconfigurable nodes to operate as a stable networking layer.

Support for New Computing Technology

The Internet "grew up" in the era of the personal computer, and has co-evolved to support that mode of computing. The PC is a mature technology today, and from that perspective, so is the Internet. But in 10 years, computing is going to look very different. Historically, when computing was expensive, many users shared one computer -- a pattern of "many to one". As computing got cheaper, we got the personal computer -- one computer per person. There was convenience and simplicity in the "one to one" ratio, and we have "stuck at one" for almost 20 years. But as computing continues to get cheaper, we are entering a new era, when we get "unstuck from one", and we have many computers to one person. We see the start of this transition, and the pace of change will be rapid. We can expect to be surrounded by many computing devices, supporting processing, human interfaces, storage, communications and so on. All these must be networked together, must be able to discover each other, and configure themselves into larger systems as appropriate.

In 10 years, most of the computers we deploy will not resemble PCs, they will be small sensors and actuators in buildings, cars, and the environment, to monitor health, traffic, weather, pollution, science experiments, surveillance, military undertakings, and so on. Today, prototypes of these computers are not hooked directly to the Internet but to dedicated "sensor nets", which are designed to meet the special needs of these small, specialized computers. A sensor net may in turn be hooked to the Internet for remote access, but the Internet is not addressing any of the special needs of these computers. It would seem odd if in 10 years we were still living with an Internet that did not take into account the needs of the majority of the computers then deployed. We should rethink now what we need to do to support the dominant computing paradigm 10 years from now. This will be of direct benefit to science, to the military, and to the citizen.

Sensor nets may seem very simple, and indeed because they are low-cost they avoid unjustified generality for application-specific features. But this technological simplicity and specificity does not mean that they do not have important architectural requirements. Sensors often have intermittent duty cycles, so they do not conform to the traditional end-to-end connectivity model of the classic Internet. Their design is driven by a structure that is data driven, rather than "connectivity driven". Some applications require a low and predictable latency to implement robust sense-evaluate-actuate cycles. A range of considerations such as these should be factored in to a Future Internet.

We identify the following specific design challenges in supporting new computing technology:

  • A Future Internet must take account of the specialized device networks that will support future computing devices, which will imply such architectural requirements as intermittent connectivity, data-driven communication, support of location-aware applications, and application-tuned performance.
  • A Future Internet should make it possible to extend a given sensor application across the core of the Internet, to bridge two parts of a sensor net that are part of a common sensing application but partitioned at the level of the sensor net.

New Distributed Applications and Systems

New networking and computing technologies provide an unprecedented opportunity to deliver a new generation of distributed services to end-users. The convergence of communication and computation, and its extension to all corners of the planet down to the smallest embedded device, will enable us to provide users a set of services anytime anywhere, invisibly configured across the available hardware. The key enabling factor to these new services is programmability at every level -- the ability for new software capabilities to self-configure themselves out over the network.

Today, we are seeing the first steps towards this future, where rich multimedia person to person communication is the norm rather than the exception; where every user becomes both a content publisher and a content consumer with information easily at our fingertips yet with digital rights protected; where the combined power of end host systems enables whole new paradigms of parallel computation and communication; and where the myriad of intelligent devices in our homes and offices become invisible agents on our behalf, rather than just another thing that breaks for no apparent reason and with no apparent fix.

Although the precise structure of these new applications and services may seem nebulous today, enabling their discovery is likely to be one of the most profound achievements of GENI. A common, reliable infrastructure can enable the research community to set its sights higher, rather than having to reinvent the wheel. Perhaps the best example of this is the history of networking research itself. When the first packet switched networks were developed, the intended target application was to support remote login by scientists to computing centers around the country. The Web wasn't on the radar, but the Web would have been much more difficult to invent without the Internet.

One design challenge is to understand how to build these new distributed services and applications. Engineering robust, secure, and flexible distributed systems is every bit as complex and difficult as engineering robust, secure, and flexible network protocols. Without a way to manage this complexity, both networks and distributed systems end up being fragile, insecure, and poorly suited to user needs. And like networks, models for managing this complexity can only be validated by building systems for real use on real hardware.

Another design challenge is how the Future Internet needs to adapt to support this new generation of distributed services and applications. The basic data carriage model of the current Internet is end-to-end two-party interaction. Early Internet applications grew up with just this form: two computers talking to each other -- remote login or a file transfer between two machines. But applications of today are not that simple. They are built using servers and services that are distributed around the network. The web takes advantage of proxies and mirrors, and email depends on POP and SMTP servers. There is a rich context for these servers -- they are operated by different parties, often as part of a commercial relationship; they are positioned around the network in a way that exploits locality and variation in network performance; and they stand in different trust relationships with the end-users -- some may be fully trusted and some (such as devices to carry out wiretap) have interests that are adverse to those of the users.

The original Internet design does not really acknowledge this complexity in application design. In fact, the Internet provides little support for application and service designers, and leaves to them much more of a design challenge than is appropriate. Today's more complex applications would benefit from a richer and more advanced set of application-support features. The Internet provides no information about location or performance -- any application that needs this information must work it out for itself, which leads to lots of repetitive monitoring traffic (e.g. PING). The Internet reveals nothing about cost -- if there is distance sensitive pricing, there is no online way for the application to determine this and optimize against it.

Similarly, the current Internet is conceptualized at the level of packets and end-points. Both the low-level addresses and the Domain Name System identify physical machines. But most users do not think in terms of machines. They think in terms of higher-level entities, such as information objects and people. The Web is perhaps the best example of a system for creating, storing and retrieving information objects, and applications such as email or instant messaging capture both information and people in their design. But none of these applications require, as a fundamental requirement, that one user concern himself with what specific computer is hosting one of these higher-level entities.

As a part of a Future Internet, we should include architectural considerations at these higher levels: should people have identities that cross application boundaries? What are the right sorts of names for information objects? How can we find objects if the name does not specify the location? There are many such questions to be asked and answered. But perhaps the more basic question is: once we propose answers to questions at this higher level of conceptualization, is the service interface of the current Internet (end-to-end two-party interactions) the right foundation for these higher level concepts, or will a Future Internet have a different set of lower-level services once we recognize the real needs of the higher levels?

We identify the following specific design challenges in designing new network services and applications:

  • A Future Internet must include a new set of abstractions for managing the complexity of distributed services that can scale across the planet and down to the smallest device, in a robust, secure, and flexible fashion. This must include an architecture or framework that captures and expresses an "information-centric" view of what users do.
  • A Future Internet must identify specific monitoring and control information that should be revealed to the application designer, and include the specification and interfaces to these features. For example, the Future Internet might reveal some suitable measure of expected throughput and latency between specified points.
  • A Future Internet should include a coherent design for the various name-spaces in which people are named. This design should be derived from a socio-technical analysis of different design options and their implications. There must be a justification of what sort of identification is needed at different levels, from the packet to the application.

Service in Times of Crisis

The Internet has grown up from its initial public sector funding to be a creature of the private sector, and this has happened at a time when in most countries the governments are deregulating their telecommunications operators. As a result, the services and functions Internet offers are driven by private sector priorities. A great deal of attention has been paid to better security in support of e-commerce, but much less to social needs. A very important example of a collective social need is service in times of crisis. For most consumers, of course, their access to the Internet is not even designed to stay up when the power goes down, so a disaster renders the Internet useless today. On the other hand, the Internet has tremendous potential as a tool for citizen access to information, emergency notification and to provide access to emergency services. The telephone system provides E911, and newer services such as reverse 911. These were conceived and designed in an era when voice was the only mode of communication. What could the strategies be for a multi-media network like the Internet? Could a Future Internet tell citizens of a tsunami or a tornado, based on their location? Could a Future Internet provide reliable and trustworthy information during a terrorist attack? There is tremendous potential here, but it will not happen in any organized way unless it is designed and implemented. This sort of public-sector social requirement should be a first-order goal for a Future Internet.

Much of the work on supporting citizens in times of crisis is done within the social sciences. This is another opportunity to reach out to other parts of NSF as a part of this project.

We identify the following specific design challenges in defining services for times of crisis:

  • A Future Internet should be able to allocate its resources to critical tasks while it is under attack and some of its resources have failed. (For example, it should support some analog of priority telephone access that is provided today.)
  • A Future Internet should allow users to obtain information of known authority in a timely way during times of crisis. The network (and its associated applications) should limit opportunities for flooding, fraudulent and counterfeit mis-information, and denial of service.
  • The Future Internet should allow users to obtain critical information based on their location, and request assistance based on their location.

Network Management

The term "management" describes the tasks that network operators perform, including network configuration and upgrades, monitoring operational status, and fault diagnosis and repair. The original design of the Internet did not fully take into account the need for management, and today this task is difficult and imperfect, and demands high levels of staffing, and high skill levels for those staff.

Network management is not just a problem for commercial Internet Service Providers. Any consumer who has tried to hook up a home network, only to have it fail to function, and has faced the frustration of not knowing what to do, has seen the limits of Internet management. Management, at the user level, is part of usability, and usability is a key to further penetration of the Internet into the user base. And corporations and institutions -- any organization that runs Internet technology -- suffer from the same management problems. The problem is endemic, and intellectually very hard to solve.

Better management tools are also vital to the goal of better availability. It has been estimated that perhaps 30% of network outages today are due to operator error. We cannot build a truly available network unless we deal with the problem of management.

A more sophisticated approach to management may depend on more powerful automated agents to support human decision-making. This is an opportunity to include researchers in artificial intelligence and machine learning as a part of this project.

We identify the following specific design challenges in improving network mangaement:

  • A Future Internet should allow a network operator to describe and configure his region using high-level declarations of policy, and automatic tools should configure the individual devices to conform.
  • A Future Internet should give users that detect a problem tools that diagnoses the problem, gives meaningful feedback, and reports the error to the responsible party, across the network as necessary.
  • A Future Internet should provide a way for all devices to report failures.

Economic Well-being of the Internet

The Internet has evolved from its roots as a government-funded research project to a commercial offering from the private sector. Internet Service Providers, or ISPs, provide the basic packet carriage service on which all the other services and applications in the Internet depend. The early designers of the Internet may not have fully understood this, but technical design choices can have a profound impact on industry structure. (For example, the routing protocol that connects different ISPs together, BGP, allows certain patterns of interconnection and the expression of certain business policies. An early alternative was much more restrictive, and would have only worked if there was a single monopoly provider.) Any redesign of the Internet needs to consider how to encourage progress -- the ongoing ability of industry to accommodate new advances while providing reliable service to customers.

Importantly, there are issues lurking in the current industry structure that presents barriers to progress. Two important issues are the commoditization of the open IP interface and interconnection among ISPs. The open IP interface implies that anyone, not just the ISP, can offer services and applications over the Internet. This openness has been a great driver of innovation, but the ISP may not necessarily benefit from this innovation. If all they do is carry packets, competition may drive the price of ISP service to the point where the ISP revenues do not justify upgrades and expansion. This tension can be seen today most clearly in the case of residential broadband. It also underlies the trends away from total openness to a world in which the ISPs block certain applications, and try to reserve to themselves the right to offer others. Problems of this sort have led to recent FCC intervention in the Internet.

Interconnection will always raise issues, because the ISPs that must interconnect may also be fierce competitors. In the traditional telephone carriers, problems of interconnection proved so difficult that regulators define the rules. So far, this aspect of the Internet has avoided regulation, but the problems are real. Whenever a new service, such as end-to-end quality of service, requires ISPs to negotiate jointly about how to offer and price the service, that new service may not happen.

It is very hard for a set of companies positioned within an industrial structure to collectively shift that structure. But if we can conceive of a slightly different structure that removes some of the current impairments, this may be a powerful inducement to adapt our ideas to the betterment of both users and the industry serving those users. This is an area where we plan to encourage participation in our effort from other disciplines, such as economics and business.

We identify the following specific design challenges in providing the right economic incentives:

  • A Future Internet should support routing protocols that are able to deal with the range of business policies that ISPs want to express. Issues to be considered include signaling the direction of value flow, provisioning and accounting for higher-level services, dynamic pricing, explicit distance-sensitive pricing, and alternatives to the simple interconnection models of peering and transit.
  • A Future Internet must provide a means to link the long-term resource provisioning problems at one level to the short-term resource utilization decisions (e.g. routing) at higher levels.