X-Git-Url: https://git.adam-barratt.org.uk/?a=blobdiff_plain;f=input%2Fhardware-wishlist.creole;h=c6bd8c12642cd05e9efe5b73d0759af8de9b506b;hb=49bb6faf4d049dc689d5eee19ec24ef79fbb0cd8;hp=d51e172e6d5acf0179029e29457f6140549bbd55;hpb=a54d6f35ace58d1f2631b7899bfadf0ff827bebe;p=mirror%2Fdsa-wiki.git diff --git a/input/hardware-wishlist.creole b/input/hardware-wishlist.creole index d51e172..c6bd8c1 100644 --- a/input/hardware-wishlist.creole +++ b/input/hardware-wishlist.creole @@ -1,166 +1,14 @@ = Hardware Wishlist = -== snapshot.debian.org == +== new klecker (2018) == -Currently snapshot.debian.net is operated by a single individual on their -hardware at home. It is a service archiving old binary and source packages. -Access to old packages, which have in the meantime been deleted from the -regular debian archive, allows Developers and Users to debug upgrade problems, -to check when regressions were introduced, to check if old packages had been -miscompiled, to downgrade to older versions while bugs are being fixed etc. +klecker is getting old and low on storage. It also serves too many purposes. +We would like to replace it with two 1U boxes with 10GigE network, 10TB of +storage (after RAID), and possibly a VPN endpoint for access to out of band +management. -Requirements: +== new bytemark (2019?) == -* Two systems with sufficient storage, hosted somewhere not in the US (so we - can import old non-US into it). -* Storage should be at least on the order of 8T (currently snapshot.d.n is - using about 4T), and easily expandable -* remote management stuff +Probably want to move from a blade enclosure and MSA to traditional +rack-mounted servers. Details TBD. -Proposers: - -* Joerg Jaspert/ftpteam -* Peter Palfrader - - -== data.d.o == - -A service by ftpmaster to host larger arch-all packages for datasets like -scientific databases (e.g. RCSB PDB, a database of protein structures) or -datasets for games. - -This service could (and probably should) share hardware with snapshot. - -Proposers: - -* ftpteam/Joerg Jaspert - - - -== merge.d.o == - -I'd like to use Ubuntu's merge-o-matic to generate diffs between Debian's -source archive and the source archives of various other Debian-based distros -(Knoppix, Freespire, Mepis, Sidux, gNewSense and so on). The result will be -much like patches.ubuntu.com. merge-o-matic downloads source packages, unpacks -them and generates diffs against unpacked pure debian source packages. As a -result lots of disk space would be required since the whole Debian archive plus -an unpacked version of it and the same for each derivative distribution is -needed. - -Requirements: - -* a system with sufficient disk space (how much is that?) - -Proposers: - -* Paul Wise - - -== source.debian.org == - -Idea: A machine which has all sources extracted from orig.tar.gz + diff applied for all dists. - -Requirements: - -* a system with sufficient disk space (how much is that?) - -Proposers: - -* Noel Koethe - -== new ftp-master == - -ftp-master's hardware is becoming old and warranty is running out (we keep extending it, but that's not for free either). - -We probably should look into getting a new machine somewhere in the US. Apart -from recent CPUs, reasonable amount of ram (16-32g?) and the usual management -fu primary requirement is reliable storage. We probably should look at some -raid6, either internal or external for the master copy of the archive in the range -of 2-4T (is that right, ftp folks?). Additionally some faster internal storage -could be useful for the database part of the ftp archive (4 disks raid10?). - -== new backup.d.o == - -bartok, the current backup.d.o, is getting old too, and there is not all that -much spare space. - -We should be looking at replacing that eventually. A raid6 of big SATA disks -would probably provide the required robustness and be cost-effective. - - -= Historical = - -These proposals probably are no longer relevant: - -== new debian mail setup == - -A set of systems that will handle all incoming mail for all debian systems. -Currently our incoming mail handling is on the individual host hosting a -service, i.e. on master.debian.org for @debian.org, on bugs.debian.org for the -bug tracking system, on lists.debian.org for our mailinglists and on several -other systems for their individual, smaller email traffic. - -Centralizing email handling will allow us to maintain our anti-spam measures in -a single point, avoiding duplicate work and hopefully improving our success. - -Requirements: - -* Four or so systems, in at least two different locations, capable of handling - modern anti-spam software. This probably needs a bit of CPU. -* remote management stuff - -Proposers: - -* Martin Zobel-Helas -* Stephen Gran - -We probably have sufficient hardware for this. Current plan involves using one -new box from HP, murphy, a new old sparc that zobel gets from some place, and -puccini that will soon no longer have packages on it. - -Status (2009-05-02): - -We probably still eventually want to move to a more cenralized setup for all -the low-traffic leaf-sites, but momentum on the Big Mail Setup Change seems -to have pretty much died. Since Stephen put quite a lot of work into making -our exim setup more readable, maintainable and we are using the same config now -everywhere, at least some of the reasons for this proposal are no longer valid. -It still might make sense to eventually move @debian.org mail from master to a new -system but we don't need 4 dedicated hosts for that, probably. (weasel) - -== new bugs.d.o == - -Currently bugs runs on a single DL385g1 system which cannot keep up with the -load that the BTS causes. - -Owner@bugs would like to split the BTS accross multiple hosts: two for incoming -email and spam filtering (would not be required if we had the setup mentioned -above), one master, and at least two user-facing web servers. - -Requirements (assuming the above mentioned mail system is in place, else add -two mail servers): - -* two systems with fast disks (we don't need that much storage, some 200 gigs - should suffice easily for a while - say 4x140 gig raid10), some ram for - caching (say 16g?), and the CPU to handle the scripts (if we can get two quad - cores per box that would be great). -* one master that processes incoming email, changing bugs as required, and - pushes the changes to the web facing servers. - -Proposers: - -* Don Armstrong - -If the snapshot hosts go through we might be able to put the bugs front end -webservers on them too. Probably a question of load but it can't hardly be -worse than rietz at the moment. - -Status (2009-05-04): - -We were quite successful in using other hosts as bugs web mirrors (even if -right now we aren't running any due to other hardware failing), so that is -a more or less solved problem. We also intent to move bugs-master to a kvm -domain at ubc/ece once the blade there is fixed. And we can probably move -incoming MX to a blade instance in darmstadt and another one at the same -place as new bugs-master will be if the bugs folks still want that.