Le vendredi 26 juillet 2013 à 13:32 +0000, "Jóhann B. Guðmundsson" a écrit : > On 07/26/2013 01:07 PM, Michael Scherer wrote: > > > > Working in IT @Red Hat, I concur, and I am pretty sure that no one has > > all the information to make that estimation. Network, hosting and > > storage are all under different budgets for different team, and all > > aggregated ( cause the DC where RH host Fedora server is not dedicated > > to Fedora, far from it ), and everybody has better things to do that > > splitting usage by project, > > Not following what you mean by project. Fedora is a project, but so does gluster, ovirt, jboss, etc. Some of them are not hosted at all by RH, or hosted externally with RH people that serve as sysadmin, either officially, fully or not fully. And some are in the same DC than Fedora. I can hardly give more details, since that's handled by a totally different departement than mine, and each community still have lots of freedom on what to choose, but for example, jboss.org, who is hosting lots of RH-sponsored project, is hosted in Phoenix, as a quick mtr show. And each project has lots of freedom, the requirement of Fedora team are not the same as jboss ( ie, Fedora would never accept jira as a bug tracker ). So by project, i mean "upstream project where RH sponsored the infrastructure, fully or not fully, with hardware or not and with people or not". > On one hand you have Red Hat the company and on the other you have > Fedora the project which means two entirely separated infrastructures. > yeah sure these two might be communicating heavily between themselves > unless ofcourse you want to risk issues from either the company or the > project being able to directly affect each other when someone screws up > or something fails or something needs to be updated in either the > company or the project and that's just bad administrator practices. They are separated on a network level (different vlans, maybe different switchs/rack, depending on the space constraint) and on organisational level but likely not in different rooms and for sure not in different datacenters for efficiency reasons ( ie, handling less datacenters is more efficient than having to deal with 5 or 6 of them ). That's the main reason to have the sponsored servers in the same place for different projects, with RH taking care of paying the local IT person when there is a problem. Please also note there is lots of servers used by Fedora that are located where no one from RH has physical access, as seen on the list of DC used by Fedora infra : https://fedoraproject.org/wiki/Infrastructure/Architecture Having served as a sysadmin for a distribution project in the past, you always see a tension between the need to have a process to grant access based on merit, technical knowledge and trust, and the harsh reality of having to pay to get to the DC when it is not located where your contributors live ( in our case, every jump to the DC costed ~ 400€ and 1 or 2 days of vacations days, due to DC being far away and opened only during weeks ). -- Michael Scherer -- devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxxx https://admin.fedoraproject.org/mailman/listinfo/devel
