high CPU usage with TAR files

[Matt McCutchen]

On 10/9/07, James V <jverge at hotmail.com> wrote:
> I am trying to use rsync to backup our Linux server and when the rsync is
> running it uses 1-2% CPU for most operations until it reaches a tar file (or
> another file archive), then the CPU usage increases to 90%.  I am assuming
> that rsync is looking at the individual files in the tar archive and seeing
> if any have changed instead of looking at the mod time and file size.

No, rsync does not do that; it treats all regular files as opaque.  A
more likely explanation for the CPU load is that rsync detected that
the tar file as a whole had changed and was delta-transferring the
contents, which can take a lot of CPU on the sending machine.  If CPU
time is more of a problem than network bandwidth, you can disable the
delta-transfer algorithm by passing --whole-file.

Matt