The TOTAL layer uses a token to implement totally-ordered delivery. Before a message can be sent, a member has to acquire the token. The member then places a sequence number on each message it sends, and messages that arrive out of sequence are delayed until they can be delivered in order. Token requests and token passing is done as much as possible on existing traffic. In fact, if the traffic load is high, all requests and token passes are piggybacked, resulting in extremely efficient totally ordered communication. The token is not fixed in one spot (sequencer), nor is it cycled through all members. Instead, it cycles through the current set of senders, an approach that works well if a set of senders present a uniform, high load. If, on the other hand, the traffic load is low, the latency and overhead of our protocol becomes relatively high. For these applications we are developing a second totally ordered layer that will use a different strategy. As with other layers, applications will be able to choose at run-time which layer they require.