I use Tobi Oetiker's Smokeping. I traceroute into major networks, and find the first IP that's pingable, and make a graphing/alarming target out of that. This helps to test latency and loss from my network into each of those networks at the nearest major NAP. Outbound packets from my own transit carriers will enter into each of those other backbones at the nearest major NAP, regardless of the destination location, and performance from my network into each of those carriers is critical. Each of those peering sessions carries 5+% of my outbound packets.
If your hosting is regional in focus, you can also pick a set of businesses in your market area, with one target singly-homed to each major carrier, and ping them. This helps you spot problems with packets getting into your own transit carriers' networks in your market, or at your nearest major NAP.
If your focus is national or global, then you would want to pick a set of targets, with one target that is singly-homed to each major carrier's network in each major market. For example, if you buy transit from MegaNet, SuperCarrier and BigTier in Dallas, but not TierZero, then you ping someone in Miami who is single-homed to TierZero, to verify that packets are getting from TierZero into one of your transit provires, without problems, in the Miami market.
If you ping two or three destinations on 10 carrier networks in 20 different markets, that gives you a pretty good idea of who is sucking wind, where.
