kutz@bgsuvax.UUCP (Kenneth J. Kutz) (12/11/90)
We are in the process of evaluating X capable devices. Below is a brief summary of the cost/benefit tradeoffs in comparing alternatives. At our site, this acquisition is being viewed as an upgrade to a dumb terminal lab. Each box represents a "level" of software that we can choose to pull off of the host system and put into the X display station. With each level comes an associated cost. The four options (and their associated software levels are): OPTIONS: (1) (2) (3) (4) +-------------+ | Desktop SW | +-------------+ +-------------+ | Window Mgr | | Window Mgr | +-------------+ +-------------+ +-------------+ | X Server | | X Server | | X Server | +-------------+ +-------------+ +-------------+ INCREMENTAL COST PER DEVICE OVER PREVIOUS OPTION: +$500 +$750-$1000 +$1500 Option (1) is provided by GraphOn, a solution that would not work for us. Our host systems will already be in need of upgrades with any of the remaining options so putting the X Server on the host (and making sure there is software loaded on each host for this task) is not an optimal solution. Option (2) is the typical X Terminal (e.g. NCD 15b, 16, 19b) configuration. This option, although better than option (1) still requires the window manager to run on a host somewhere presenting the following problems: (2a) MULTIPLE POINTS OF FAILURE: if the host running the window manager crashes, with it goes your WM capabilities on your X terminal. Recovery from such a crash can be clumsy. (2b) WINDOW MANAGEMENT PERFORMANCE DEPENDENT ON HOST PERFORMANCE: moving windows around on the screen can become tedious with a window manager running on a host already over burdened (2c) WM MOUSE EVENTS TRAVEL OVER THE NETWORK RATHER THAN STAYING LOCAL The marketers have distinguished the device of option 3 as an X Station rather than an X terminal. Given option (3) alleviates (2a), (2b), and (2c), the cost of these benefits may well be worth it. Recently option (4) has popped up in our discussions here. By "Desktop SW" I mean software which supports a desktop environment similar to the Mac, OpenWindows (free), X.desktop or Looking Glass. For the average user, this environment eliminates the need for having to learn the operating system commands. Option (4) then would be a low cost diskless workstation (such as a Sun SLC) which would run all three levels of software locally rather than burdening the host with any of those tasks. This last option however is radically different from the first 3 for the following reasons: (4a) THE EXECUTABLE IMAGE AND THE DATA ARE NOW ON DIFFERENT MACHINES (i.e. all data access is now via NFS). This is significant for commands such as "grep strings.h *.c" in a populated source directory. (4b) THE DISK HOME FOR THE EXECUTABLE IMAGE AND CPU TO RUN THAT IMAGE ARE ON DIFFERENT MACHINES (i.e. typing 'ls' on a diskless workstation generates NFS traffic in the form of the 'ls' executable traveling to the node which is non-existent in options 1-3) (4c) KERBEROS SECURITY PROBLEMS: unencrypted tickets and passwords may be paged (and thus available) over the network (4d) MULTIPLE POINTS OF FAILURE: losing an NFS server is more serious than losing a machine running your window manager as your entire display becomes unusable at this point. Although not mentioned above, option (5) might be option (4) + a local SCSI disk to eliminate problem (4c) as well as reducing network traffic from option (4). (NOTE: Because this equipment purchase is a replacement for a dumb terminal lab, we have purposely left out the obvious benefit of local processing availability for options (4) and (5). It is assumed for this acquisition that the computing itself (compilations, editing, statistical analysis, etc.) will not be done on the desktop. The only local processing in view here has to do with X, window and desktop management, all of which are independent of our user's application) o What we are interested in is any quantitative or subjective studies that have been done on the performance of the entire system (from display device to network to host) when one compares option (3) with option (4). o How much of an impact would an NFS accelerator board (such as the one offered by Sun through a third party) have on our decision between options (3) and (4)? o Has anyone successfully implemented option (4) to provide their users with a common desktop user interface ACROSS VARIOUS OPERATING SYSTEMS for command line functions such as renaming, copying, removing, and editing files? Comments on any or all of these issue would be appreciated. I will post a summary of the responses (not merely a concatenation) for those who are interested. -- Kenneth J. Kutz Internet kutz@andy.bgsu.edu Systems Programmer BITNET KUTZ@ANDY University Computer Services UUCP ...!osu-cis!bgsuvax!kutz Bowling Green State Univ. US Mail 238 Math Science, BG OH 43403