chl@cs.man.ac.uk (Charles Lindsey) (02/15/90)
We are looking into the possibility of using an Encore Multimax as a dedicated fileserver for a collection of Sun workstations. I would be interested to hear from people who have experience of this sort of application, and also to have views as to whether the sort of thing I am proposing below is practcable, sensible, or whatever. Replies by email, but I could summarise any interesting conclusions to the net. The setup we have here consists of about 200 workstations, nearly all discless, and mostly 3/50s, but with some 3/60s, 3/80s and SPARCs. These are served by a great variety of servers (3/180, 4/280, 4/110, and even SPARCstations). As the system grows, it gets ever more difficult to manage, so a large central server holding exactly one (and preferably the only) copy of each thing seems a good idea, although we would not expect to throw away all the existing servers at once, but rather to move gradually to a more rational system. And we would not permit anything but NFS traffic on the Encore, since we have plenty of other machines to use as compute servers when needed. Our chief worry is that most of the traffic to/from our discless clients is swapping, so my first question is whether anyone out there allows any of their workstations to swap directly to the Encore, and can quantify the load that this imposes? For example, how many discless stations does it take to swamp one ethernet segment, and what fraction of an Encore cpu is needed per workstation? Also, is there any feel for how much memory to buy per cpu (if indeed it is a linear relationship at all)? (Of course, there is always the alternative strategy of having local servers to do just swapping duty, but it is always fun to imagine the worst-case scenario first!) My second question concerns whether system management would really be radically simplified. I have been trying to imagine a scenario in which we have a new version of Sun OS to put up, so we clear out a whole disc on the Multimax, nominate one client as the guinea pig, and install the new OS on it and that disc. Then, when it is seen to be working OK, we just have to reorganise the mount tables for the other clients and they all find themselves in the new regime. Hopefully, one did not have to dump or relocate any user filestore (always one of the major hassles at the moment with a new Sun OS release), and one can even have different groups of machines remaining at different OS levels. Is it really as simple as all that, or am I being too naive? And finally, since we would be booting the Suns off the Multimax, are there any thrashing problems or whatever if you suddenly order a massive reboot of all the workstations on the system (conventional Sun servers have been known to lie down and die when subjected to such indignity :-) )? My thanks in advance to anybody who can give me some advice over these problems.