POP Interconnection Fabrics

Martin Schulman, Juniper Networks

	Though there are a wide range of differences among high-speed Internet
backbone architectures, one concept they share is the "POP".  Originating
in the days of deregulation when inter-exchange carriers terminated their
long haul fiber optic cable runs at a "Point Of Presence" in a city, it now
refers to any building or room in which a service provider locates one or
more pieces of communications equipment.  Examples include telephone
company Central Offices and strategically located computer room space
leased to carriers.

	POPs usually employ a variety of equipment types.  Backbone routers
interconnect POPs using high-speed WDM, SONET, and/or ATM links.
Aggregation routers connect large numbers of customers using dial, ISDN,
xDSL, T1, and other technologies.  Sometimes colocated servers provide
management, security, or data storage functions for scalability.  These
devices are often from different vendors, and display widely varying
traffic and serviceability needs.

	The earliest POPs employed FDDI, and some have transitioned to the more
economical Fast Ethernet.  In 1998 the industry began to deploy OC12-based
routers, and in 1999 many are expected to upgrade to OC48.  While multiple
parallel FE links could accommodate some of the load, OC48s will require an
unreasonable number of parallel 100 Mbps links.  Thus, the industry must
upgrade the fabrics, and some will consider next generation technologies.

	This paper will examine the requirements of the POP interconnection
fabric - bandwidth, topology, scalability, etc.  Existing solutions including
ATM, SONET, and Gigabit Ethernet will be compared to the industry needs.
Emphasis will be placed on understanding the problem space and the reasons
that motivate a carrier to choose a given solution.