Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On 12/1/06, Igor Yu. Zhbanov <bsg at uniyar.ac.ru> wrote: > Hello! > > Is it possible o build SAN with DRBD v0.8 and some (which one?) cluster file > system? > > Suppose, we have two client nodes (which will access our SAN) and two storage > nodes. All nodes connectet by, suppose, gigabit ethernet. On storage nodes > we will run DRBD v0.8 in Primary/Primary mode, so we can access to our hard > drives on both nodes simultaneously to make load balancing (at least for > reading). > > Next, I think, we need to set up some clustered file system upon our DRBD > devices pair. Probably it will be GFS or Lustre (What is the best? Also, > I don't know is it possible to setup Lustre in Primary/Primary configuration). > > So, we can mount file system on both nodes. It's all ok. Both storage nodes > can mount file system and use it in parallel. But what about two our clients > nodes? Is it possible to mount GFS or Lustre or something else remoutely? > Or am I must to setup NFS (which nobody likes) on top of GFS? > > Please, tell me your suggestions, is it possible to build load-balancing SAN > with parallel access to each storage nodes and multiple clients nodes? > (And is it possible without exporting shared block device to All nodes which > want to access to shared file system? I think, network file system trafic is > much lesser than network block device trafic.) > > > Thanks! > I believe Lustre requires a minimum of 4 nodes. (2 for Metadata serving and 2 for object serving. See the drawing at: http://lustre.org/docs/Lustre-Datasheet-v1.4-06.pdf Is that acceptable to you? (I don't know if you can have your clients running on any of those 4 nodes or not.) == Quote from http://lustre.org/docs/selecting-a-cfs.pdf Lustre High Availability Lustre failover delivers completely application transparent system call completion. Lustre Metadata Servers are configured in an active/passive pair, while Object Storage Servers are typically deployed in an active/active configuration that provides redundancy without extra overhead. == End quote So it seems reasonable that the Object Storage servers could use drdb v0.8 to provide redundancy. You might ask on the OpenSSI list. They support both drbd v0.7 and Lustre so maybe they could answer if it makes since to try the 2 together. Greg -- Greg Freemyer The Norcross Group Forensics for the 21st Century