Rollup is the process by which the core server removes unneeded data. When a base image reaches the end of a retention policy, it is rolled up into the next incremental so that the base image is no longer outside the retention policy. If there is more than one base image for a machine, the oldest base image will eventually be deleted when it is no longer inside the retention policy and there are no incremental recovery points that rely on it.
If the retention policy is set so that the oldest recovery points kept are daily recovery points (and no weekly, monthly, or yearly recovery points are kept) then the base image will have to be rolled up each day. If the oldest recovery points kept are weekly recovery points, then the base image will be rolled up each week. When combined with large base images, this retention policy configuration results in very large and time consuming rollup jobs running on the core each day or each week.
To prevent the base image from being rolled up each day or week, set the retention policy so that at least 1 monthly or yearly recovery point is kept. This configuration may result in more data being kept in the repository, but it is generally a small quantity since it is only one recovery point.