Throttling gsutil


Particularly when used with the -m (multi-threading) option, gsutil can consume a significant amount of network bandwidth. In some cases this can cause problems, for example if you start a large rsync operation over a network link that's also used by a number of other important jobs.

While gsutil has no built-in support for throttling requests, there are several strategies, as well as various tools available on Linux and macOS that can be used to throttle gsutil requests. If you typically use the -m option, the first step to reducing your network bandwidth is simply to remove -m from your commands.

If you want to further throttle your requests, one tool is trickle (available via apt-get on Ubuntu systems), which lets you limit how much bandwidth gsutil consumes. For example, the following command limits upload and download bandwidth consumed by gsutil rsync to 100 KBps:

trickle -d 100 -u 100 gsutil -o "GSUtil:parallel_process_count=1" \
  -o "GSUtil:parallel_thread_count=1" rsync -r ./dir gs://some bucket

NOTE: We recommend against using the -m flag with gsutil when running via trickle, as this may cause resource starvation and prevent your command from finishing.

Another tool is ionice (built in to many Linux systems), which lets you limit how much I/O capacity gsutil consumes (e.g., to avoid letting it monopolize your local disk). For example, the following command reduces I/O priority of gsutil so it doesn't monopolize your local disk:

ionice -c 2 -n 7 gsutil -m rsync -r ./dir gs://some bucket

Another way to adjust your bandwidth consumption is to modify the values for parallel_process_count and parallel_thread_count. These parameters are set in your .boto configuration file, but also can be controlled on a per-command basis using the top-level -o option.