walltime

Walltime, short for wall clock time, is the real elapsed time between the start and end of a task as perceived by a clock in the outside world. It encompasses all delays caused by reading from or writing to storage, waiting for resources, scheduling decisions, and any other external factors. This makes walltime different from CPU time, which measures only the amount of processor time consumed by a process.

In practice, walltime is what a user or a scheduler experiences as the duration of a job. It is commonly reported as “real” or “elapsed” time in benchmarking tools, in contrast to CPU time, which is divided into user time (time spent executing user code) and system time (time spent in kernel code on behalf of the process). Walltime can be longer than CPU time, especially for I/O-bound or multi-threaded workloads that wait on resources or synchronization.

Scheduling systems, such as batch queues and high-performance computing schedulers, often enforce a walltime limit on jobs. Users request a certain amount of walltime, and when the limit is reached, the system may terminate the job or send a signal to wrap up. This policy helps allocate resources fairly but can lead to incomplete tasks if the limit is set too low.

When measuring performance, walltime can be influenced by changes to the system clock. For precise timing that is robust to clock adjustments, monotonic clocks or other elapsed-time APIs are recommended, particularly in benchmarking or long-running measurements.

See also: elapsed time, real time, latency, CPU time.