Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
Jason Gray a écrit : > I'm looking at creating a multi-clustered array server network for our > production environment. Is it possible to have 5,6,7..n servers clustered > together (kind of like a token ring) to act as redundant arrays for each > other? drbd is peer-to-peer. As long as you have one secondary assigned each primary, drbd will consider different volumes as independently managed, so you can build some kind of a ring, where each server backs-up the previous one. > > So, instead of having a Primary array I would have 5,6 or 7 "Primary" arrays > that mirror each other across an isolated network (10.0.0.0 say). each primary volume is mirrored only once. It cannot "broadcast" updates to others, but only sends them to its secondary. I can see a major advantage in your proposal, because your P volumes will be small and will resynch faster. A computer failure will only affect a smaller amount of data and less users. > Each server would provide a work space for 10-20 users. If the server they > are working on goes down they are routed to a different server. The whole > server network acts like a 5, 6 or 7 server RAID cluster. This is failover. It has to be done by another piece of software. Heartbeat is an off-the-shelf solution for two servers. I manage the Shaman-X project which works out a kit for N servers and P drbd volumes. It does what I understand you ask for. the current version is there: http://downloads.shaman-x.org/recovery_kits/shx-kit-17-may-04.tar download it on a workstation and follow read-me instructions. I take this opportunity to provide project status update below. This kit is an interative programme called cdx in Shaman-X terminology which prompts you for your parameters and then generates a custom-kit to be deployed on your n servers. Running the custom-kit then requires wdx, sendarp, drbd, php and apache on every servers. Sendarp comes with the .tar above. WDX can be downloaded at http://downloads.shaman-x.org/wdx-0.4.2.src.tar Management of the whole is done through a web page called hdx, so you know at anytime which is primary and which is secondary. Good points are: - It manages N servers and P drbd volumes - Primaries and secondaries are both failed over. I mean, if a secondary fails, a new secondary is chozen and re-synchs. - You can assign primary/secondary roles manually for every drbd volume from hdx web page Shaman-X team is all volunteers and therefore we lack time for finalizing, so following limitations apply now: - cdx custom-kit generator is not tested enough and there are probably still a few bugs in it (shell script). - hdx web page is planned to be re-engineered (php script): bug corrections, display improvements, password protection for commands and documentation. Today for instance, HDX needs to declare two separate computer rooms for disaster protection, when this should be an option. - custom kit should be improved: today, if the secondary is not fully synchronized, it accepts to switch to primary status, which is bad - Network failover is not implemented is to be implemented in the custom-kit too (relocate primary and secondary according to network status) - the custom kit implementation of drbd we made is to be checked by a drbd specialist, and intensive testing is necessary before going to production. - I hope I'm not discouraging you. It is closer to a good result than it appears above. Testing is certainly the key activity to plan. > > The issue is that the secondary systems are essentially offline. Can the > systems write to each other while being online? I need each node to be > active and sending data to each other. Is this possible? No it is not. The secondary volume is off-line. A server X can be primary for volume 3, 4 and 5 and secondary for 1 and 7. It will serve volumes 3, 4 and 5 and support IP3, IP4 and IP5 addresses used to access data. Volumes 1 and 2 can be served by server Y which is secondary for volumes 3, 4 and 6. Y will support IP1 and IP2. Volumes 6 and 7 is served by server Z which is secondary for volumes 5 and 2. Z will support IP6 and IP7 When a server fails, for all primary volumes move the primary roles to the corresponding secondaries, for all secondary volumes, choose a new secondary and resynch from the primary. > > > Cheers, Hope this helps Regards Dominique > > Jason > > _______________________________________________ > drbd-user mailing list > drbd-user at lists.linbit.com > http://lists.linbit.com/mailman/listinfo/drbd-user > >