bobp@tekcae.TEK.COM (Robert N. Perry (Bob)) (09/23/87)
Here's a few questions for all you cluster freaks out there: 1) How does one designate a node as being a compute server on a cluster? 2) How does one designate a node as abeing a disk server? 3) What is the function of the cluster password in an LAVC ? 4) Does everyone who tries to log in to a cluster have to know the cluster password? 5) How do terminal servers do load shedding in an LAVC? I've heard it's done by only looking at the number of users logged in, not by actual system/cpu load. 6) Anyone heard any rumors about license enforcement on clusters by DEC in future releases of VMS (to prohibit non-licensed cluster nodes from running images licensed on the boot node only)? -- Robert N. Perry (Bob) Tektronix-Beaverton, Oregon "Skydive for better health" bobp%tekcae@tektronix.TEK.COM "When I works, I works hard. When I sits, I sits easy. When I thinks, I goes to sleep."
SYSTEM@CRNLNS.BITNET (09/26/87)
Bob, > Here's a few questions for all you cluster freaks out there: > >1) How does one designate a node as being a compute server on a cluster? One doesn't. Although if a cpu had no local disks, you would want to discourage i/o bound jobs somehow. One can, however, distribute compute loads by deciding which nodes have batch queues and varying the various queue limits appropriately (job size, runtime, number of simultaneous jobs, etc). Anybody on any cluster node can submit a batch job to queues running on any other node in the cluster (if the queue protections allow). You can also set up "generic" queues to feed jobs to the first free corresponding cpu-specific queues. >2) How does one designate a node as being a disk server? In a "real" (CI) cluster, HSC50s and HSC70s are disk servers. Alternatively, a disk server is any cpu with a lot of disks. At boot time, you tell SYSGEN to load the MSCP server, and then you do a "MOUNT/CLUSTER" to make the appropriate disks available to the rest of the cpus in the cluster. In a LAVc, a disk server is also any cpu with a lot of disks. You run AUTOCONFIG with "MSCP_LOAD = 1". Then, at boot time, you do a "MOUNT/CLUSTER" to make the appropriate disks available to the rest of the cpus in the cluster. This is the default for the "boot-node(s)". (This is the case for VMS 4.6. I believe 4.5 was a bit more restrictive.) One can set up a common authorization file to make a user's default directory be on a specific disk on any specific cpu in the cluster, no matter what cpu is used for login; or you can set it up so the local disk is always used. There is more than enough flexibility. If you use logical names, you could move the disk from one cpu to another (between terminal sessions) and the user need never know. >3) What is the function of the cluster password in an LAVC ? I believe it is used at boot time to verify that a cpu is joining the right cluster. >4) Does everyone who tries to log in to a cluster have to know the cluster >password? No. It's only used internally by the operating system. >5) How do terminal servers do load shedding in an LAVC? I've heard it's >done by only looking at the number of users logged in, not by actual >system/cpu load. The algorithm is more complicated. (We don't use it. We let people log into whatever cpu they want. We don't have enough yet for it to make a difference. All of our systems are "fully loaded") >6) Anyone heard any rumors about license enforcement on clusters by DEC in >future releases of VMS (to prohibit non-licensed cluster nodes from running >images licensed on the boot node only)? only rumors. I hope this helps. Selden E. Ball, Jr. (Wilson Lab's network and system manager) Cornell University NYNEX: +1-607-255-0688 Laboratory of Nuclear Studies BITNET: SYSTEM@CRNLNS Wilson Synchrotron Lab ARPA: SYSTEM%CRNLNS.BITNET@WISCVM.WISC.EDU Judd Falls & Dryden Road HEPnet/SPAN: Ithaca, NY, USA 14853 LNS61::SYSTEM = 44283::SYSTEM (node 43.251)