wytten@umn-cs.cs.umn.edu (Dale Wyttenbach) (06/07/90)
> > Were there any other important issues that influenced your decision? Our main reason for buying the SLC's probably falls under the versatility category. If we bought X-terminals, all the X clients would have to be run on some existing CS dept CPU - almost certainly umn-cs since people prefer it. umn-cs is a 6 processor Sequent S-27 with 32MB of RAM and 140MB of swap space. Right now, umn-cs runs out of virtual memory when we get about 70 users on, and the active processes include say 20 xterms, 10 emacs, 10 xloads, 10 xbiffs, etc. We need to educate users to run X clients on their local workstation instead of umn-cs. For example, if everyone used 'xterm -e rlogin umn-cs' instead of 'rcmd umn-cs xterm' it would help a lot. Obviously people can't run clients locally if they have an X-terminal on their desk. One of the most basic resources we provide is CPU cycles. For about the same price as either an NCD-19 or a Visual-19, you can add a lot of CPU cycles to your pool by buying an SLC instead. Who knows how these CPU cycles will be used 3 years from now, but they'll be available. Comparing xbench results for the SLC vs. the NCD-19, the SLC won some battles and lost others. However, the NCD server is highly optimized, and Xsun is not. If and when Xsun is improved, I suspect that the SLC would clobber existing X-terminals. dale - Dale Wyttenbach | ...rutgers!umn-cs!wytten wytten@cs.umn.edu | wytten@umnacvx.bitnet Computer Science Department Systems Staff--University of Minnesota, Minneapolis
de5@STC06.CTD.ORNL.GOV (SILL D E) (06/07/90)
In article <1990Jun7.144309.29740@cs.umn.edu>, wytten@umn-cs.cs.umn.edu (Dale Wyttenbach) writes: > >Our main reason for buying the SLC's probably falls under the >versatility category. Diskless workstations *are* more versatile, and in your case they were probably the right way to go. Others who are less constrained by server horsepower and perhaps more constrained by network loading and per-user cost would probably be better off with at least some percentage of X terminals. I used to think diskless workstations had it all over X terminals-- until I ran a diskless SPARCstation off my own SPARCstation. The administrative overhead(1), server burden(2), and increased network throughput(3) have convinced me that that's not the way to go. There are other considerations, too. Workstation users have to be a little more sophisticated, in general. Someone needs to administer the workstation, and you need to consider the problem of controlling root access. (1) setting up and maintaining diskful-server/diskless-client pairs is substantially harder than hanging an X terminal off the Ethernet. (2) my response time on the server was often frustratingly impacted by the diskless client. Contrast that with the NCD-16 I also used--it never had a noticeable impact on the systems that served it, beyond that inherent in the work being done. (3) the NFS traffic on the Ethernet caused by the diskless box accessing files--data and binaries--and swapping was sufficient to cause clients *I* was running to hang/die. -- Dave Sill (de5@ornl.gov) These are my opinions. Martin Marietta Energy Systems Workstation Support