International Engineering Consortium
Web ProForums
Carrier Service-Level Agreements (SLAs)

4. What Service Levels, by Components, Should Be Negotiated with the Service Provider?

The service-level–negotiation process should be performed in a careful manner with much attention to detail. Most carriers demand a minimum number of sites (usually ten) before they enter into negotiation of service-level guarantees. Nearly all service providers exclude certain components from their SLA measurements. These exclusions may be negotiable. The following guidelines can be applied to the negotiation to ensure a successful implementation of the frame-relay network.

Network Availability

Most carriers are committing to a monthly availability guarantee of at least 99.5 percent. The availability guarantee includes all components of the carrier frame relay network, the carrier-provided local loop to the frame relay network, and the carrier-provided CSU/DSU (see Figure 3).


Figure 3. Availability Guarantee

Excluded items may include the following:

  • customer CSU/DSU, router, FRAD, or other access device
  • customer-provided local loop to the frame-relay network
  • network downtime caused by the carrier's scheduled maintenance window
  • customer-induced outages
  • backup or dial PVCs
  • acts of God
  • Note the distinction between network-based and site-based availability. For a ten-site network, in a thirty-day month, a 99.5 percent average network availability would allow thirty-six total hours of downtime (not including the exclusions mentioned above). If the SLA is written around a site-based availability instead of being network-based, then any one site can only be down for 3.6 hours in the month. This distinction is very important when computing downtime. Often, availability SLAs that are network- or site-based do not meet the business requirements, so PVC availability is a better metric.

    PVC Availability

    It is better to commit your carriers to a monthly PVC availability. Typically guarantees are 99.5 percent. This has the desired impact of restricting the amount of downtime any one PVC will sustain. PVC availability is critical if a company is utilizing a frame-relay network to carry delay and application-sensitive traffic such as SNA or voice. If SNA sessions are lost, reestablishing them could take longer if the frame-relay outage resulted in front end or application timeouts. Companies should make sure their carrier provides the PVC availability their specific application traffic requires. Included and excluded components are identical to those of network availability.

    Round-Trip Network and PVC Delay

    A wide range of delay guarantees are available and are dependent upon the carriers' specific configuration guidelines and backbone technology. Round-trip delay guarantees range from less than 110 milliseconds to nearly 300 milliseconds. Some carriers provide guarantees based on access-line speed, offering much lower delay guarantees for T1 access lines than for 56 kbps. However, the most useful delay guarantees are those based on the type of application or type of traffic. SNA, voice, and video traffic are much more sensitive to round trip delays than FTP or HTTP TCP/IP sessions. Guarantees specifically targeted at those application types are critical to the success of a frame-relay implementation. Table 1 illustrates an example of an application-specific guarantee. (Companies should work with carriers to understand their specific offerings and the requirements to gain the lowest committed round-trip delay.)

    Additionally, some carriers expect the customer to be highly involved in the measurement process—often to the extent of the customer being held responsible for initiating packet Internet groper (PING) tests in concert with carrier technicians during times of light network utilization. Also, round-trip network delay often is based on predetermined frame sizes (200 to 250 bytes). Companies should understand how the process works during negotiations. An IP PING test is not recommended, because it is an inaccurate measurement of frame-relay delay. There are two reasons for this inaccuracy; first, IP PING has low network priority, and second, IP PING–delay measurements include router delay.

    The best test of network delay is based on frame-relay frames, independent of the router, and measures consistently throughout the day. Traffic load and delay should be measured when its impact is at its highest—peak traffic load—with this information collected and processed for trending analysis. Included and excluded components are identical to those of network availability.

    Throughput

    A wide range of throughput guarantees is available and is dependent upon the carriers' specific configuration guidelines and backbone technology. Carrier guarantees range from 99 percent to 99.999 percent, dependent on the carrier and their specific guarantee language. Some carriers base the percentage of delivered frames on committed information rate (CIR) or if the frames are labeled "discard eligible (DE)." Others base the calculation on committed burst size versus excess burst size (uncommitted data in excess of committed burst size). Guarantees are therefore difficult to compare without significant study and understanding of the carrier's specific terminology. Carriers often require a minimum number of PVCs per link (often four or more). In addition, carriers exclude configurations where it is determined that the egress (destination) port has not been configured with enough bandwidth or CIR. However, the means of verifying this fact is undefined. Strong process language must be in place to ensure proper guarantee execution.

    Additionally, companies must ensure that the carrier has redundant means of gathering SLA data. Some carriers exclude lost data from throughput calculations. Other excluded items may include the following:

  • data loss during the carrier's scheduled maintenance window
  • any month in which the customer fails to transmit a predetermined amount of data
  • backup PVCs
  • dial PVCs
  • PVCs or access channels added or reconfigured during the month
  • Mean Time to Respond

    The mean time to respond is usually four hours if the site is within a predetermined number of miles from a carrier-maintenance center. Longer timeframes are determined by distance from that center. Maintenance in this instance covers only the carrier-provided CSU/DSU and local-access facilities.

    Mean Time to Repair/Restore

    The mean time to repair/restore is usually four hours. Note that response and restoral totals result in an aggregated outage time of eight hours.

    It is clear that when negotiating an SLA with the frame-relay carrier, careful study and understanding of their terminology, exclusions, processes, and calculation methodology is critical.

    Registered Users
    Enjoy exclusive access to free On-Line Education and receive the biweekly IEC newsletter.

    IEC Newsletter
    Get the latest industry information including critical insights from key industry leaders, technology briefings, and an Analyst Corner.
    Current
    Subscribe

    Newsroom

    IEC Corporate Member

    Advertising Kit