On Tue, Apr 15, 2008 at 07:07:37PM +0200, Dr. Peter Poeml wrote:
On Tue, Apr 15, 2008 at 04:31:54PM +0200, Detlef Steuer wrote:
May be it's enough to stop mirroring (but keeping!) end-of-live products plus introducing some kind of quota system?
I see two aspects here, one is diskspace on download.opensuse.org, another is diskspace on the mirrors.
The former is a decision done elsewhere, let's see what opensuse-project thinks about it, the suggestion to move the discussion there is a good idea.
The latter is something that we easily can take care about, with a little bit of preparation.
I just gathered some data. What they show is the following:
- 10.1 distro usage: It is not downloaded in any significant amount, we can definitely remove it from all rsync modules right now (or rather move into a separate "discontinued" rsync module so it is actually still available). We actually could have done so earlier, and even though no mirror so far complained about the size of those rsync modules, it is always good to keep them small because in the end it might enable more mirrors.
- 10.1 BS repositories usage: The popular repos (KDE:, server:, OpenOffice.org, and others) are all accessed with about 1 req/s. The overall average traffic generated looks like only 25 K/s. It should nicely work to exclude all those from the rsync modules, thereby shrinking them and removing them from the mirrors, without impact on our download infrastructure. [Overall request rate for BS repos is 80 req/s, 1 req/s of that is for 10.1 repos if I don't mistake anything.]
I gathered some more (build service specific) data, which may be of interest. The following numbers are space occupied on rsync.opensuse.org by repositories matching the following expressions: *Mandriva_2006* 1G *Mandriva_2007* 2G *Mandriva_2008* 2G *Debian* 0.5G *Ubuntu* 1G *Fedora_7* 5G *Fedora_8* 8G *RHEL_4* 0.1G *RHEL_5* 2G *SLES_9* 9G *SLE_10* 31G *10.0* 19G *10.1* 45G *10.2* 73G *10.3* 88G *Factory* 100G TOTAL 393G TOTAL w/o home 223G So, these numbers allow for some interesting insight. However they'll be fully useful together with statistics about downloads, and I'll wait for But I _seriously_ wonder already now, whether we need to have 100G of Factory rpms on the mirrors, which are probably downloaded only by extremely few users. A quick look into the logs indicates that the requests on the 100G Factory packages cause _less_ of traffic for us, as the requests on the 10.1 packages... Peter -- "WARNING: This bug is visible to non-employees. Please be respectful!" SUSE LINUX Products GmbH Research & Development