bob@morningstar.com (Bob Sutterfield) (11/16/90)
We have four SS1s on a network. Two are diskless, two serve user filesystems, and of those two, one also serves / and /export to the other three. All four run big stuff (X11 and various GNU tools) and are fairly heavily used as development machines. Right now, two have 12M and two have 8M. Suppose we have the opportunity to increase our total pool of memory so that two have 16M and two have 12M. Which two should get the 16Mb? One school of thought would hold that the machines that serve files are more globally critical, and should have more memory for nfsd(8)s so as to better service the demand. Another school contends, just as plausibly, that the diskless machines should have more memory so that they present less paging load to the network and the diskful machine that holds their swapfiles. (One might, in a more cynical moment, note that the two positions are each held by an equal number of people with SS1s on their desk; and one might further note that the position held by a given person corresponds very closely to the presence or absence of a local disk attached to that person's SS1.) (These two positions might be related to economic cure proposals put forth by various political parties: one suggests that we reduce the demand, the other that we increase the supply. Similarly to our network's residents, the position held depends largely upon whether one is on the supply side or the demand side.) Are there any studies, white papers, rules-of-thumb, collected wisdom, or oral folklore that address this issue? In other words, would anyone like to step in and break our tie? :-)