The asynchronous model of distributed systems has no bounds on

	execution entity (the principles of distributed computing community uses the term "process") execution latencies –-  i.e., arbitrarily long (but finite) times may occur between execution steps
	message transmission latencies –- i.e., a message may be received an arbitrarily long time after it was sent
	clock drift rates -- i.e., process's local clocks may be arbitrarily unsynchronized

In other words, the asynchronous distributed system model makes no assumptions about the time intervals involved in any behaviors.

The synchronous model of distributed systems, conversely, has a' priori known upper and lower bounds on these quantities

	each process has a bounded time between its execution steps
	each message is transmitted over a channel and received in a bounded time
	process’s local clocks may drift either from each other or from global physical time only by a bounded rate (note that this does not necessarily bound the clocks' accuracy)

It might seem that any implemented distributed system could be considered synchronous by presuming unrealistically tiny lower bounds and huge upper bounds on these three quantities. But the intent of having a system model is to facilitate reasoning about (e.g., proving) properties of the system -- if those presumed bounds are not deterministic (i.e., guaranteed), reasoning based on them can not be deterministic, which may or may not be acceptable in any given circumstances. In particular, meeting hard deadlines requires reasoning based on deterministic worst case latencies, and doing so for multi-node behaviors requires that the system be synchronous.

The asynchronous model is the general case, and any proofs of its properties also apply to the synchronous model, but the converse is not true. However, a number of important properties (such as distributed agreement) have been proven impossible even under very weak conditions in the asynchronous model (e.g., [Fisher et al. 85]), but are readily achievable in the synchronous model.

Neither the asynchronous nor the synchronous models are pragmatic for the vast majority of actual implemented distributed systems. Both are extreme vertices in a space of distributed system models. Distributed system models elsewhere in that space are partially synchronous along some dimension(s). Examples include (but are not limited to): a system may have either bounded process step latencies and unbounded communication latencies, or vice versa; a system may be asynchronous with synchronous subsystems; a system may change from one model to another at different times.