larryc@puente.jpl.nasa.gov (Larry Carroll) (04/30/91)
We have an Ethernet network of a bunch of DECstation 3100s running X & Motif. Currently the X is DEC's DECwindows, which is X11R3 (or so I'm told), but we're soon migrating to a generic X11R4. There's also a VAX with about a gigabyte of disc space that we'll use to store raster maps. All systems run Ultrix & are NFS'd together. We're writing a map-viewing & -annotating program which will run on the workstations. The VAX will act mostly as a file server. For reason I don't fully understand, the designers wants to allow users to launch as many copies of the map program as they choose to. They also want to use a local server to maintain working replicas of map files on the local hard disc for the map program. This local server will run as a daemon & (presumably) by started up each time the system is booted. My question is: what's the best way implement this daemon? Where can I read up on the trade-offs of various methods of implementing it? Or does NFS have some way of automatically maintaining these replicas the way the NCS system does? (At least the book we have on NCS suggests that, although I've not been able to find any specifics on that.) Can NFS be configured so that other disks on the network act as caches, and if so can the cache contents be made long-term? (Users will often work on a particular map over a period of days or even weeks.)
jik@athena.mit.edu (Jonathan I. Kamens) (05/01/91)
(Note: It's comp.unix.programmer, not comp.unix.programmers.) (Note2: Note the Followup-To.) The Andrew File System (AFS) does two things which seem specifically pertinent to what you're trying to do. First of all, it allows you to create multiple read-only backup volumes of a volume of files, and any access to the files in the volume will be balanced over the various backup volumes, thus distributing the load of reading the files. Second, it keeps a local cache of AFS files on every workstation. Often-used files get copied into the local cache and then stay there. Subsequent accesses to those files are as fast as accesses to files on the local disk. There have been several papers presented at various Usenix conferences about making NFS do mirroring and stuff like that. The Summary 1990 conference had at least one paper of this sort. So if you really want to do this with NFS, see if you can find someone where you work who has old proceedings from Usenix conferences, and ask to see them (or contact Usenix and ask if you can purchase relavant proceedings). However, if I were you, I'd consider using AFS instead. You can probably get the "standard blurb" about AFS by sending mail to info-afs-request@transarc.com. -- Jonathan Kamens USnail: MIT Project Athena 11 Ashford Terrace jik@Athena.MIT.EDU Allston, MA 02134 Office: 617-253-8085 Home: 617-782-0710
ian@rathe.cs.umn.edu (Ian Hogg) (05/01/91)
In article <1991Apr30.043758.2072@jpl-devvax.jpl.nasa.gov> larryc@puente.jpl.nasa.gov (Larry Carroll) writes: >We have an Ethernet network of a bunch of DECstation 3100s running X & >Motif. Currently the X is DEC's DECwindows, which is X11R3 (or so I'm >told), but we're soon migrating to a generic X11R4. There's also a >VAX with about a gigabyte of disc space that we'll use to store raster >maps. All systems run Ultrix & are NFS'd together. > >We're writing a map-viewing & -annotating program which will run on the >workstations. The VAX will act mostly as a file server. For reason I don't >fully understand, the designers wants to allow users to launch as many >copies of the map program as they choose to. They also want to use a local >server to maintain working replicas of map files on the local hard disc for >the map program. This local server will run as a daemon & (presumably) by >started up each time the system is booted. > >My question is: what's the best way implement this daemon? Where can I read >up on the trade-offs of various methods of implementing it? Or does NFS have >some way of automatically maintaining these replicas the way the NCS system >does? (At least the book we have on NCS suggests that, although I've not >been able to find any specifics on that.) Can NFS be configured so that >other disks on the network act as caches, and if so can the cache contents be >made long-term? (Users will often work on a particular map over a period of >days or even weeks.) I think you are confused about NFS and NCS. NFS is a network file system as you know. NCS is not a network file system. It stands for Network Computing System and is a type of RPC technology. NFS machines communicate with each using Sun RPC's. The replication available in NCS is called Data Replication Manager (DRM) and only guarantees that all servers of a particular class will eventually see all the calls made to the class of servers. I believe that it is not fully implemented. It is only implemented in the Location Broker. Some file systems (like AFS) support file replication. Your daemon implementation depends on how the application uses the map files. If only one user can access the file in write mode, why not just keep a list of checked out files on the VAX and have each node copy the file to their local machine when needed? -- Ian Hogg email: rathe!ian@cs.umn.edu ...!umn-cs!rathe!ian Rathe, Inc ianhogg@cs.umn.edu 366 Jackson Street phone: (612) 225-1401