On 4/16/07, Jonathan Dieter <jdieter@xxxxxxxxx> wrote:
Someone else asked about the format of the diff, etc. Deltarpms can be combined quite easily. For example you might have a deltarpm for foo-1.0=>1.1 and a deltarpm for foo-1.1=>1.2. You can combine the two into foo-1.0=>1.2 and it will be roughly the same size as if you ran makedeltarpm foo-1.0 foo-1.2. This means that we don't *have* to keep old rpms around to create deltarpms. We just have to have the latest rpm for each package + old drpms. I will write whatever needs to be written to make pruning work, etc. I'm just not quite there yet.
That is great that the diffs can be combined in this way. It allows a lot of flexibility in how the source lays out the diffs for the mirrors to clone. I guess my point, now that I know that presto is able to do this, would be that while the result of a merging of two incremental diffs won't be much different in size than doing the diffs directly, the keeping the two diffs separate themselves will be much larger than a single diff bridging the same version gap; especially if the package changes are done to the same files repeatedly. So I guess the trade-off has to be made as to whether you keep around a lot of diffs and use up some storage space just to accommodate giving some efficiency to those who don't update frequently. Thomas suggests 14 days. I would be more inclined to give a month. In the end, if worse comes to worse, people just end up downloading the full version of the rpm so it is not the end of the world. Maybe there are some stats available from the various yum repo maintainers as to when requests for specific package updates drop off for those packages not superseded by another update. /Mike -- fedora-devel-list mailing list fedora-devel-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/fedora-devel-list
