If you've been reading about all of the updates of our servers, you know that they are now getting fairly large. Four of our file servers sit at about 20 Terabytes, and one sits at about 48 Terabytes. In addition, we're caring for another 48 Terabyte server of Ed Fox. And Confocal data will wind up sitting on a third 48 Tb server. Further, we are looking in the face of several more next generation sequencers. We have been working with Nikos George and Terry Haley to try to create a more unified, consistent feeling for our users.
Today I created a great bunch of NFS-mounts on RESCOMP1, which is an 8-processor, dual-core server. All partitions on the RESSRVxx servers and all partitions on the RCBIGxx servers are mounted to RESCOMP1. If there is sufficient throughput, we should be able to have all users login through RESCOMP1. Their home directory will be on an NFS-mounted partition, but the actual server should be irrelevant to our users. Ideally, you would connect by ssh or netatalk or samba to RESCOMP1 and your files could be, essentially, anywhere, connected by the network, or iSCSI, or fiber-channel. They would all appear local to RESCOMP1. (In an emergency, if RESCOMP1 were to fail, you could still reach your files by know which server they are really on.)
While we are testing, users will continue to connect to their server shares as always.
We're going to set up some test cases, make sure shared file areas, whether via AFP or Samba, work correctly. We hope to have users migrating to RESCOMP1 over the next ... we're not sure how long it will take, exactly. And we don't know, yet, given the big variety of user file types, how many users can, realistically, be located on one home machine. We might need a second.
we also need top consider the nature of our storage and its scalability. While we are doing this, we're working with Terry Haley (form Ed Fox's Lab) to have a real research LDAP authentication server, which could authenticate, define "home" directories, and groups for the broader Research community. And then we have to figure out the relationships among the various centers. It's a rather large issue.