Distributed Systems Group
[ Project areas ] [ Overview ] [ Members ] [ Publications ] [ Presentations ]
TRIAD is a new next generation Internet Architecture, which
defines an explicit content layer that provides scalable
content routing, caching, content transformation and load balancing,
integrating naming, routing and transport connection setup.
TRIAD is an acronym, standing for
Translating Relaying Internet Architecture integrating
The Internet Name Resolution Protocol (DRP) and Name-Based Routing Protocol (NBRP) integrate routing, naming, and connection setup.
The Wide-Area Relaying Protocol (WRAP) is a shim protocol which provides extended addressing and improved loose-source routing. WRAP also uses name-based TCP checksums to provide end-to-end checks even in the presence of network address translation. WRAP can be used for VPNs (Virtual Private Networks) or policy-based routing as well.
Middleware support for multi-source application takes the form of application-level relaying. Relay nodes in the network send packets from multiple sources over a common EXPRESS channel, providing floor control support and other services.
With the emergence of the web, the primary use of the Internet is content distribution, i.e. web pages, and increasingly audio and video streams. Some measurements indicate that 70 to 80 percent of Internet traffic is HTTP traffic (and most of the rest of the traffic is routing and DNS). That is, almost all of the traffic in the wide area is delivery of content, and ancillary traffic to locate it and determine how to deliver it. Today, millions of clients are accessing thousands of web sites on a daily basis, with the top 20 web sites supplying about 10 percent of the content. Moreover, new popular web sites and temporarily attractive web sites can prompt the arrival of a so-called "flash crowd" of clients, often overwhelming the resources of the associated web site.
To scale content delivery to support these demands, a variety of ad hoc and, in some cases, proprietary mechanisms have been deployed. For instance, content is geographically replicated at multiple sites with specialized name servers that redirect DNS lookups to nearby (to the client) sites based on specialized routing, load monitoring and Internet "mapping" mechanisms, so-called content routing. Further, proxies in the network provide transparent caching of content, further reducing the load on web sites and the network resources between the cache and the web site. Finally, load-balancing switches at each web site allow the virtual host for the web site to be realized as N physical hosts, distributing the content requests across these hosts based on individual server load and the content it holds. Future extensions are expected to provide content transformations proxies that will transform content to a representation suitable for the requesting client, such as a mobile PDA.
These mechanisms violate the original Internet architecture in various ways, do not fully address scalable content distribution, and compromise the basic philosophy of the Internet as being based on open, community-based standards. In particular, content routing requires a DNS server that accesses some form of routing information, a layer violation (or duplication of the routing layer), yet still requires the world-wide clients of a site to access a single centralized server as part of accessing that site. This roundtrip to a central server is quickly becoming the dominant performance issue for clients as Internet data rates move to multiple gigabits, reducing the transfer time for content to insignificance. Transparent caching requires hijacking transport-level connections, violating the end-to-end semantics and causing connection failures when the routing changes to route around the cache. Moreover, these caches introduce extra delay in connection setup for non-cachable content while they collect the HTTP header information required to determine whether the content is cachable or not. Load balancing switches rely on network address translation (NAT) which requires rewriting addresses, port numbers, transport-layer checksums and even packet data, making mockery of the original end-to-end semantics. (Network Address Translation is also being widely deployed for other reasons, such as address allocation autonomy, multi-homing, concealing endpoints and increasing the number of addresses available in an edge network.
In 1992, work on IPng, next-generation IP, was initiated based on concern that the Internet was running out of addresses. This effort, predating the web, focused on providing more addresses but failed to anticipate the strong emergence of content as the primary use of the web and thus failed to address the content distribution issues described above.
In particular, all end-to-end identification in TRIAD is based on names and
URLs, with IP addresses reduced to the role of transient routing tags.
At this content layer,
the integrated directory, routing and connection setup to provide
efficient content routing to replicated content
and eliminate roundtrip times to access the content in most cases.
Finally, TRIAD supports path-based addressing using a simple ``shim'' protocol
on top of IPv4 called WRAP, for content routing control
and extensible addressing.
This extensible addressing eliminates the need to transition the Internet
We claim that TRIAD not only provides scalable content distribution,
but also solves the Internet problems associated with
network address translation.
TRIAD also provides attractive solutions to mobility,
virtual private networking, policy-based routing and source spoofing.
TRIAD can be incrementally deployed,
initially without changes to end-hosts
or applications beyond that already required for NAT.
Professor David Cheriton
Mark Gritter and David R. Cheriton:
An Architecture for Content Routing Support in the Internet.
In the USENIX Symposium on Internet Technologies and Systems, March 2001
[ html ]
[ ps.gz ]
David R. Cheriton and Mark Gritter: TRIAD: a Scalable Deployable NAT-based Internet Architecture Technical Report, January 2000. [ ps.gz ]
Katerina Argyraki and David R. Cheriton:
Loose Source Routing as a Mechanism for Traffic Policies.
In the ACM Workshop on Future Directions in Network Architecture, August 2004
[ pdf ]
An Architecture for Content Routing Support in the Internet,
USITS presentation, March 26, 2001.
Mark Gritter: Naming and routing in TRIAD, Computer Forum poster slides, 2001.
Mark Gritter: The TRIAD Content Layer: An Internet Architecture for Content Routing Support, Stanford networking seminar, November 2, 2000.
Vince Laviano: Scalable Multi-Source Multicast Sessions in a Single-Source Multicast Network, Stanford networking seminar.
Chetan Rai: Wide-area Relay Addressing Protocol (WRAP): Packet Delivery in TRIAD, Stanford networking seminar, October 28, 1999.