Signaling
Media and router performance are increasing very rapidly, so one may conclude that the time of infinite throughput is at hand. However, such is not the case. Signaling could pose a serious impediment to full use of the bandwidth that media and routing/switching will make available.
Signaling refers to the process by which an end system notifies a network that it wants service. The network responds with resources (bandwidth, buffering, and entries in databases) that allow the connection to proceed, or it notifies the end system that resources are not available. At the end of a connection, signaling is used to release resources. During a session, signaling is used to change attributes, for example, adding bandwidth, a third-party, or capabilities such as access to services.
The most familiar example of signaling is picking up a telephone and dialing a phone number. A computer at the telephone company notices when you pick up the phone. Upon receiving the phone number, a route is established to your requested destination, and bandwidth is reserved. If you dial a 500-, 800-, 888-, or 900-number, there is a database lookup to find the "real" phone number that is hidden from the caller, and a route is established to the phone number found in the database. Recent innovations, such as the implementation of packet network in the telephone network, enable all this signaling to occur much more quickly than previously. Even so, the call setup process is time-consuming and expensive. Modern switches are capable of only a few hundred call setups per second.
Signaling requires a round trip from the origination point of the network to the destination to determine whether sufficient resources are available at that instant in time. A round trip goes through the Access Network and the Core Network, during which lots of things happen: Switches reserve resources such as memory, data bases are consulted, transmission links are checked for bandwidth availability, and end systems are validated for proper addressing.
As an alternative to the significant housekeeping chores of telephone signaling, connectionless networking emerged as a highly scalable networking architecture. The Internet is the paradigm of connectionless networking.
Connectionless protocols such as IP impose a minimal signaling burden on data communications equipment and end user devices because they do not have a call setup process. If resources are available, the packet arrives. If not, the packet is discarded. No call setup takes place. It's easier to ask for forgiveness than for permission.
Signaling protocols exist for all networks. In addition to the telephone example cited previously, examples of signaling protocols are Resource Reservation Protocol (RSVP), for IP networks such as the Internet; ATM Q.2931 which establishes call setups for ATM; and various flow setup protocols for packet switching.
Viewpoint: Signaling Rate as an RBB BottleneckWhen one hears of RBB networks, one generally thinks of the requirement for bandwidth and speed. However a strong argument can be made that the fundamental technical bottleneck for RBB is not bandwidth but signaling rate. To take an extreme example for illustration, if a telephone switch were capable of only one call setup per minute, then it wouldn't matter that the switch can transmit 1 Gbps of information. The throughput would never be realized because so few callers could use the switch (of course, no one can talk that fast anyway). The same holds for data switching equipment. If routers and ATM switches cannot support sufficiently high signaling rates, then the bandwidth offered by fiber optic network cannot be realized. Current measurements on the Internet, obtained from the ' Group North American Network Operators , show that one can expect on average less than 20 packets per data flow and that traffic is highly bimodal in terms of packet length. That is, packets are either very short (64 bytes) or very long (1500 bytes, about the maximum length of an IP packet). A large fraction of Internet packets are acknowledgments, which are of minimal length. The impact of short flow length is to increase the signaling required to sustain maximum throughput through the network. For example if an IP switch has 16 ports of OC48 (2.488 Gb) port speed each, aggregate speed of the switch is 40 Gb. If average packet size is 256 bytes, then the router must transmit 20 million packets per second. If there are 20 packets per flow, then the router must be capable of 1 million flow setups per second. This is far beyond the capability of any IP switch, ATM switch, or voice switch available today—or likely to be available in the near future. These statistics suggest that signaling rate will choke the switch long before packet forwarding is choked. These calculations are not meant to suggest that a million flow setups per second are needed for RBB. In fact, we don't know what the required flow setup rate is because the applications are not built yet. The point is simply that the bottleneck through a network may not be bandwidth, but signaling. Therefore, applications and networks should be designed with signaling rate in mind. |