FRAME RELAY: CIR and Billing Issues Revealed

Frame relay has become a popular public data service offering for users, equipment vendors, and public service providers alike since its initial availability in the U.S. in 1991. Frame relay is one member in the family of emerging fast packet technologies and services that also includes Switched Multimegabit Data Service (SMDS) and Asynchronous Transfer Mode (ATM), all of which are characterized by high speed, high throughput, high performance, low error rates, and error detection and correction by end-to-end hosts.

Frame relay, in particular, provides an unreliable virtual circuit service. It is a virtual circuit (VC) service because two end users must establish a logical connection prior to any data exchange, although the network does not actually dedicate any resources to that connection. Multiple virtual circuits may share a single physical circuit on a statistically multiplexed basis. The VC is unreliable because the network provides no guarantee of delivery nor flow control. In a frame relay network, frames with errors are discarded and frames experiencing severe congestion may be discarded, and the end user device is not notified of these events by the network; instead, error recovery is performed on an end-to-end basis. Due to the very low error rates in today's optical fiber, digital networks, the unreliable nature of frame relay (and other fast packet services) is, in fact, not a drawback.

All of today's frame relay offerings in the U.S. offer Permanent Virtual Circuit (PVC) service, where VCs are defined between user sites upon service subscription. Switched Virtual Circuit (SVC) service, which will allow the establishment of VCs on a per-call basis, should be available from some providers by the end of 1995. Because of the connection-oriented nature of frame relay, it is well-suited to centralized applications and/or those with predictable traffic patterns, such as private line replacement, transport of SNA traffic, access to the Internet, or other environments with many sites doing most of their communication with a central point.

For such a useful, simple service, however, many users are confused when discussing some aspects of frame relay. The Committed Information Rate (CIR) parameter associated with frame relay is perhaps that service's least understood feature. It is defined and packaged differently by different service providers. Increasingly, these differences are being used as a marketing tool and many are losing sight of the purpose of CIR. As a result, CIR fuels confusion rather than provide its real function — that of balancing optimal use of network bandwidth with the needs of the individual virtual circuits that share a transmission facility.

To understand CIR, it is enlightening to look at its definition according to the standards. The main function of CIR is to provide fair access to the network's bandwidth. Consider the scenario of an interactive CAD/CAM workstation and point-of-sale (POS) terminal, each with a VC across a single physical interface. If the workstation is sending large frames on a relatively continuous basis, the POS terminal's intermittent short frames would experience unfair delays.

The CIR provides a mechanism of forcing these two devices to share the bandwidth is some sort of equitable fashion. Suppose the physical interface is a DS0 (56 or 64 kbps) facility. In this case, we might assign a 48-kbps CIR to the workstation and a 16-kbps CIR to the POS terminal. This provides fair access for both stations since they each have a guaranteed (sort of!) bandwidth allocation.

But now examine this situation from a slightly different perspective. The CAD/CAM workstation has been throttled to a speed of 48 kbps even when the POS terminal is idle, meaning that 8 kbps on the access facility are unused. This doesn't make sense from the viewpoint of the workstation, physical access facility, or the network, because we don't want to waste this bandwidth. Allowing a user to transmit at a rate above the CIR, then, might have its merits in some circumstances.

So, how is CIR actually defined? It is important to understand that CIR is not an instantaneous measurement of transmission, but an average rate over time. Thus, a 16-kbps CIR does not mean that a user is limited to a transmission rate of 16 kbps, but that the user will not transmit more than 16,000 bits in a second (or 32,000 bits in two seconds, etc.). This distinction is not just a subtlety and nuance, but goes to the very heart of how CIRs work. Consider that when the POS terminal above transmits, it will send at a rate of 64 kbps because that is the access facility's bit rate; as long as the POS terminal doesn't transmit for more than 1/4 second, however, there is no violation of the 16-kbps CIR (assuming that we measure CIR over a period of 1 second).

The CIR is derived from two parameters. The first, Committed Rate Measurement Interval (Tc), is the interval of time over which information transfer rates are measured. We assumed a Tc value of 1 second above, which is the most common value; different service providers may use a different value. The second parameter, Committed Burst Size (Bc), is the maximum number of bits the network guarantees to deliver during the time interval Tc under normal circumstances.

The CIR is defined as the throughput rate, in bits per second, that the network agrees to support, under normal conditions; it is calculated by dividing the Committed Burst Size by the Committed Rate Measurement Interval (Bc/Tc).

Note that a large Tc value may be better for customers because it allows more aggregation of bandwidth over time and supports highly bursty traffic (meaning that most of the transmissions take place in a relatively short amount of time); a small value of Tc is better for the network because it minimizes burstiness. If Bc = 16,000 bits and Tc = 1 second, for example, the CIR is said to be 16 kbps. However, a Bc = 160,000 and Tc = 10 also yields a CIR of 16 kbps. The latter case, however, will allow a more erratic (bursty) traffic pattern.

Most data transmissions are, in fact, bursty in nature. If we believe in the burstiness of data and the bandwidth efficiency of statistical multiplexing, we will also believe that most user devices will not constantly operate at their CIR. If network traffic is bursty, then the network ought to accommodate one device's excess traffic if there is available bandwidth. This is desirable because if the network were to discard a given user's burst of traffic, the user would just re-submit it later.

To address this concern, most frame relay networks try to handle a momentary excess burst if there is bandwidth available. But how much excess will the network tolerate? This is defined by a third parameter called the Excess Burst Size (Be), which is the maximum number of bits above the CIR that the network will attempt to deliver during the Committed Rate Measurement Interval. There is a risk if you transmit above the CIR, however; information transmitted at a rate above the CIR is marked discard eligible by the network.

There is a maximum limit to how much information the user can submit in a given time interval. If a user transmits more than (Bc+Be) bits in time Tc, the excess frames are discarded immediately by the network. There is no standard name for this value, although the term Maximum Burst Rate is sometimes used. To avoid frame discards at the entry node of the network, Be is commonly set to be the difference between Bc and the access rate of the interface, so that (Bc+Be)/Tc equals the access speed.

To further muddy the waters, a new term is being used to specify how must bursting is allowed, namely the Excess Information Rate (EIR). This may be defined in a number of different ways, but the EIR is usually calculated as Be/Tc, with a Be value chosen so that (CIR+EIR) equals the access speed.

FIGURE 1: CIR and related parameters: Access speed, Bc, Be, Tc, and DE.

Figure 1 shows this graphically by examining the number of the bits transmitted over a period of time. The line labeled Access speed represents the line speed of the physical access channel, in bps. The CIR point is the intersection point of the horizontal Bc line (bits) and vertical Tc line (seconds).

During the early part of the measurement time interval, the user transmits two frames with a total number of bits below Bc. Since the user transmits at the line speed, the quantity of user data increases at the same slope as the Access speed line. After transmitting both frames, the total quantity of traffic is still below the CIR. In both frames, the Discard Eligible (DE) bit is off (0).

The third frame causes the user to exceed the CIR agreement. In this case, the network will accept the frame but will mark it discard eligible (DE=1). The fourth frame causes the user to exceed the excess burst agreement and is discarded immediately by the network.

With this generic description behind us, let's now take a look at some of the ways in which CIR is defined and used by the frame relay service providers. The answer is —inconsistently! And this is adding to customer's confusion about the service.

Consider the following three frame relay service offerings:

• WilTel limits the CIR of a given VC to the speed of the line. If the CIR is less than the speed of the line, the traffic rate on the VC may burst up to the line speed. But they allow the sum of the CIRs of all VCs on a single line to equal 200% of the line speed (400% in the case of DS0).

• Sprint defines a "zero CIR." That is, no VC is limited to anything other than the line speed. Of course, every transmission is in excess of the CIR, so all frames are marked discard eligible.

• Pacific Bell limits the sum total of all VC's CIRs to the line speed and does not support the excess burst feature. Consider a 64-kbps access line with 50 VCs. This suggests that each VC, then, is limited to about 1200 bps.

So what do we learn from this? Is WilTel suggesting that I can really get 256-kbps throughput on a 64-kbps line? Does Sprint want me to be happy knowing that they consider all of my data to be discard eligible? Does PacBell believe that we should build high-speed networks so that we can allocate low-speed virtual circuits?

Much more fundamentally — how does this make a difference to the end user?

It might be instructive to cut through the marketing distinctions and look at the bottom-line ramifications of these different philosophies. WilTel balances between optimal use of network bandwidth and fair access by users. Their CIR procedure assumes that incoming traffic is highly bursty and, therefore, wants each user to use all of the line's bandwidth if it is available. They also assume that all VCs will never need more speed than the line can deliver at any given point in time.

Sprint, on the other hand, assumes that the incoming traffic on all VCs has roughly the same burst characteristics and that any CIR value may needlessly throttle the traffic on some VC. Using a 0-CIR means that all VCs compete for the available bandwidth; this results in optimal bandwidth use for the network (because there is never idle time when a station wants to transmit) and, if all traffic patterns are similar, fair access for individual VCs.

PacBell appears to come down in the middle. CIRs are necessary to ensure the user data does not flood the network and false expectations are eliminated by limiting the aggregate CIR to the port speed.

Table 1 summarizes the different ways in which CIRs are used by some of the frame relay service providers in the U.S. The bottom line before buying a service is to know how CIR is defined by the provider and know what it means to you. Also, you have to understand how your frame relay customer premises equipment handles CIRs. The frame relay service providers and equipment vendors are increasingly using CIR as a marketing tool and sometimes lose sight of the real impact on the customer. Different definitions and uses of CIR causes confusion but often the differences have, in fact, limited impact on the user.

As an aside, some local exchange carriers (LECs) do not define CIR in their frame relay service tariffs; while the frames are not marked discard eligible, this is effectively the same as 0 CIR. For frame relay connections involving a hand-off of frames between the LEC and interexchange carrier (IEC), however, CIR must be defined in some manner across the Network-to-Network Interface (NNI) since all of the IECs use CIR.

While understanding CIRs is important to understanding the frame relay service (and products) that you are buying, it is not the only thing you need to know. It is also often hard to compare frame relay services costs between different providers because the tariffs seem to have different pricing components and options.

The first thing to look for is the local loop charge. The local loop is the physical connection between the frame relay equipment on your premises and the network. All frame relay services require a local loop; sometimes this charge is quoted as a separate line item and sometimes it is included in the overall quote. Local loop charges are usually distance-sensitive.

The second thing to look for is a backhaul charge, which is the cost of the connection between the network office where your local loop terminates and your service provider's nearest frame relay switch, or point-of-presence (POP). This is a particularly important issue when buying your local loop from one carrier (such as the LEC) and your frame relay service from another (such as an IEC). Again, backhaul charges may be quoted separately or bundled into the service charge, and are also usually distance-sensitive.

What gets most of our attention is the third item, or the rate for actual data exchange. There are two basic pricing models for frame relay services: flat rate and usage-sensitive.

Most frame relay service providers bill a flat monthly charge, usually based upon the port connection, CIR, and/or number of PVCs. WilTel, for example, bills their WilPak service based upon the port connection speed and the sum of the CIR of all PVCs defined at a customer's site. The CIR is measured in a simplex fashion, reflecting only the outgoing traffic (more on this later). CIRs are assigned to individual PVCs and may range from 16 kbps to the port speed, in increments of 16 kbps. The total aggregate CIR at an interface may total up to 200% of the port speed (up to 400% for 56/64 kbps ports). Local loop access and backhaul charges are additional.

MCI also bills on a fixed rate basis, but uses different criteria than WilTel. Their Hyperstream Frame Relay Service charges are based upon two components. The first is the port speed. The second component is the mileage and CIR of each individual PVC. Local loop and backhaul charges are additional.

Most local exchange carriers employ an even different cost basis. BellSouth, for example, bills each PVC on a CIR and port speed basis, while NYNEX and Southwestern Bell charges are based upon port speed and the number of PVCs.

The second frame relay service billing model is usage-based, possibly with a maximum monthly rate. MCI and Sprint both employ usage-based billing. MCI actually has two usage-based plans. In their Usage CIR PVC plan, usage charges range from 30% to 100% of the basic fixed-rate plan with the same CIR, and usage charges are based upon CIR, distance, the amount of traffic within the CIR limit, and the amount of traffic that exceeds the CIR. In their Zero CIR PVC plan, there is a small minimum monthly fee plus a usage charge; there is no monthly price cap.

There is one more wrench in the billing works. Frame relay charges are usually calculated per-access interface and based upon the number of outbound PVCs and/or the outgoing CIR. This implies that PVCs are simplex. Frame relay PVCs are, in fact, full-duplex according to the standards. Simplex PVCs are merely a convenience for billing.

There is a potential advantage for the customer with simplex billing, however, and that is the ability to define PVCs with asymmetric CIRs. AT&T and WilTel, for example, currently offer such an option.

FIGURE 2: Sample frame relay configuration showing PVCs between four customer sites; all access the network at 56 kbps and the CIR shown is that of the PVC in the user-to-network direction. Note the use of oversubscription and asymmetric PVCs.

Figure 2 shows a frame relay PVC connectivity diagram for some hypothetical customer. Table 2 summarizes the access port speeds, source and destination endpoints for the PVCs, number of PVCs at each customer site, and the sum of the CIR. Different billing schemes will use different aspects of this particular scenario. Note use of over-subscription and asymmetric PVCs. Note also that the four full-duplex PVCs are billed as eight half-duplex PVCs, because billing is done at each customer interface.

Frame relay is a popular and highly useful data service. Its simplicity is clouded somewhat by the convolutions associated with different providers' CIR definitions and billing schemes. These differences also make it difficult to compare the prices between the different service providers.

There is more information about frame relay services, products, and standards on the Internet. The Usenet comp.dcom.frame-relay newsgroup is reflected on the Internet and is an excellent way to track current discussions of relevance to frame relay and ask questions of your own.

There are also several information servers with loads of Frame Relay material. Frame Relay Forum (FRF) Implementation Agreements can be downloaded from their Web site at www.frforum.com. Motorola maintains an excellent frame relay site at www.mot.com/MIMS/ISG/tech/frame-relay/resources.html. Additional sites can be found at www.garykessler.net/library/commcomp.html#frame.