> From: Roland Mainz <roland.mainz at nrubsig.org> ... > James Carlson wrote: > > Joseph Kowalski writes: > > > However, because of all the discussion, I just want to verify the following. > > > > We also have to clarify what goes on root. I think it's at most just > > libcmd and libast, as the others seem to have nothing to do with the > > problem. (And libast could be excluded by using a lazy load, as > > nothing from root will ever reach that path.) > > libast still needs to be in /lib - if any application or script uses one > of the LD_* flags to override the lazyload flag the matching application > would be in trouble (assuming this happens before /usr - which is > unlikely but still a possibilty... ;-/ ).
I see this is a reply to my mail where the concern expressed was virtual memory space. Jim followed up with a concern about space on root. I have to say that I've not expressed too much of a concern about space on root, simply because I don't have the numbers handy to know how much space needs to be taken before we have a problem. However, I think I should point something out. These days, almost nobody worries about the cost of a few megabytes of rotating rust. It comes free in boxes of Cracker Jacks these days. Opps, there is that "almost" word. We have customers who have configured servers with lots of roots (for diskless or whatever) and they made them pretty tight. If we increase the size of root too much, we make upgrade very difficult for these people and they become unhappy customers. Like I said, I don't know the threshold here - in just outlining the senario. I think we (Solaris ON) control all the binaries that use libcmd (old flavor) before user is mounted. Hence, we could modify these utilities to not need libcmd.so (say by using a libcmd.a). We'd rather not do this, but if disk space becomes an issue, we could. - jek3
