AARNet run a comprehensive microwave network through the Australian metropolitan areas to service universities and research institutions. We rarely find RF issues on the mcirowave links themselves. The units are supplied under contract and tested bit and burst errors between the two indoor units for a day before we take acceptance. The contractors provide annual maintenance, at this time the Automatic Gain Control level is checked against recorded values and this gives indications of tree or buildings that may have appeared on the microwave path. Australia has a spectrum allocation agency that registers, licenses and taxes point-to-point microwave links. Thus there is a low incidence of interference. Australian city centers, with the exception of Sydney and Melbourne, are not as populated as in North America. Sites where the RF engineering is at the nominal performance boundaries do experience considerably more rain fade then sites that are well within those boundaries. Since there is more rain in the US than in Australia, this may be an issue for you. Because each 'customer' site does its own networking (ie: we don't engineer the network at both ends) we have had considerable difficulty with, in order: 1) getting sites to calculate G.703 electrical power levels. G.703's output power at the transmit port is unspecified in the standard, G.703's input power at the receive port is. The intent was that this allows manufacturers to design for the expected coaxial cable length. Most manufacturers have jumpers to select a "short" cable or a "long" cable. Some manufacturers require you to insert attenuation (such as a RF test resistor) instead. Some equipment, notably the Cisco LS1010 ATM switch, has differing output levels between cards with differing port densities. This complicates dealing with a port failure. If you are starting from scratch use the same card everywhere (high density, with access to the clock redistribution bus). 2) Getting sites to understand clocking. As always there should be one, and only one, clock. Preferrably generated by the upstream site. If you want to emulate a T1/E1 circuit, then the clock should also be derived from a telco clock. Unfortunately, power level and clocking misconfigurations result in the same error counters incrementing. 3) Not propoagating AIS. Most equipment will not propogate Alarm Indicatin Signal or insert RDI in their default configuration. This allows the device to be tested. Once testing is complete AIS propogation should be enabled and RDI insertion activated. Thus when a physical-layer error occurs it is instantly propogated to the router/switch interfaces at *both* ends. This is much better than relying on the routing protocol to discover the unreachability some minutes later, even if no alternative route is available (because the users instantly receive a Network Unreachable rather than timing out and the "show interface" shows the true status of the link). 4) RF engineering practices on the G.703 link between the router/switch and the microwave indoor unit. In particular: long runs parallel with non-RF cables, small turn radii, ground currents because shields are terminated at both ends on long-distance hauls. If you are running the G.703 link some distance I would seriously consider using a G.703 opitcal modem and running multimode fiber rather than coaxial cable. 5) Poor RF engineering by switch/router vendors. Thus the huge baluns on the Cisco-supplied coax cable and the warning not to manufacture your own (which, realistically, you *should* do as loops of excess G.703 coax is asking for trouble). -- Glen Turner Network Engineer (08) 8303 3936 Australian Academic and Research Network glen.turner@aarnet.edu.au http://www.aarnet.edu.au/ -- The revolution will not be televised, it will be digitised