As you may have noticed, we have seen a pretty major regression in service on mirrors.dotsrc.org. We have had a few ideas of what could cause this, and initially it looked like scrapping was the culprit.
After turning of the indexing on nginx (sorry), we really did not see much of a difference. Instead we have seen pretty stable levels of outgoing traffic, but at a much much lower speed per download.
Instead this seems to be an issue with our zfs array degrading. We think we have narrowed it down to a single very old vdev. Next week and the following weeks we will try to find a replace the worst performing disks.
We will probably take mirrors.dotsrc.org offline while this is going on, given that the service is currently almost unusable. We will try to keep this to danish night-times.