geoff@hinode.East.Sun.COM (Geoff Arnold @ Sun BOS - R.H. coast near the top) (08/09/89)
One of the questions I am often asked about PC-NFS is "how come there's no way for me to find out when a particular file server is going down? Unix users get notified." I point out that (at least on SunOS) the mechanism used is "rwall", which is an RPC service, and that for size reasons we can't afford to embed a version of rpc.rwalld in PC-NFS. This explanation is reasonable, but unsatisfactory. My reaction was to say "let's ask the NIC for a UDP port so that we can use it to send unsolicited messages to PCs running PC-NFS." That would certainly do the trick. However, a moment's thought reveals that the problem is bigger than just PC-NFS. Surprisingly, there is at present no simple ubiquitous message protocol to fulfil this function. rwall is fine for SunOS and other ONC licensees, but what about other systems? Do I have to rely upon SMTP? That's incompatible with the idea of broadcasting a simple message such as "The backbone will be down for five minutes at 12:00 to replace a bridge." This could be trivially simple or slightly more involved (but still simple). The trivial approach is to dedicate a UDP port for unsolicited system messages. Anyone could send one, in a single datagram, and the listener process would be responsible for delivering it as seemed appropriate for the system (dialog box, console message, etc.) A more complete approach would be to define a formal protocol so that it would be possible to convey information about the coding of the message, message length (so that TCP could be used instead) and so forth. [If the spec exceeds one page, it's too complicated.] Comments? Geoff Geoff Arnold, Internet: geoff@East.Sun.COM PCDS Group, Sun Microsystems Inc. --------------------------------------------------------------------------- Just think: If Unix had been developed in England, we'd all be using BCPL...
beame@maccs.dcss.mcmaster.ca (Carl Beame) (08/11/89)
In article <681@east.East.Sun.COM> geoff@hinode.East.Sun.COM (Geoff Arnold @ Sun BOS - R.H. coast near the top) writes: >One of the questions I am often asked about PC-NFS is "how come >there's no way for me to find out when a particular file server >is going down? Unix users get notified." I point out that (at least >on SunOS) the mechanism used is "rwall", which is an RPC service, and that >for size reasons we can't afford to embed a version of rpc.rwalld in >PC-NFS. This explanation is reasonable, but unsatisfactory. > > >Comments? > Since we implemented a rwalld in our BWNFS (PC based NFS client), I checked to see how much memory it takes. Not counting the initialization code, (which is not resident in memory), it takes 192 (decimal) bytes. But I guess if you write in C it takes a lot more :-) - Carl Beame Beame@McMaster.CA P.S: Geoff, does the new PC-NFS come with a fixed version of rpc.lockd for Sun OS4.0.1 ?
geoff@hinode.East.Sun.COM (Geoff Arnold @ Sun BOS - R.H. coast near the top) (08/11/89)
In article <1989Aug10.170537.1823@maccs.dcss.mcmaster.ca> beame@maccs.dcss.mcmaster.ca (Carl Beame) writes: >Since we implemented a rwalld in our BWNFS (PC based NFS client), I checked >to see how much memory it takes. Not counting the initialization code, (which >is not resident in memory), it takes 192 (decimal) bytes. I presume you're not including the portmapper in this total? We don't run a portmapper in PC-NFS, since we don't normally run any RPC based servers on the PC. (After all, we don't want to undercut our workstation business :^) "rwall" uses a pmap_rmtcall to contact rpc.rwalld, but to be correct you have to handle both direct and indirect calls, don't you? Also, are you doing the any duplicate filtering? >But I guess if you write in C it takes a lot more :-) I'm sure you know that none of the resident PC-NFS code is written in C... Geoff Arnold, Internet: geoff@East.Sun.COM PCDS Group, Sun Microsystems Inc. --------------------------------------------------------------------------- Just think: If Unix had been developed in England, we'd all be using BCPL...
beame@maccs.dcss.mcmaster.ca (Carl Beame) (08/13/89)
In article <693@east.East.Sun.COM> geoff@hinode.East.Sun.COM (Geoff Arnold @ Sun BOS - R.H. coast near the top) writes: > >I presume you're not including the portmapper in this total? We don't >run a portmapper in PC-NFS, since we don't normally run any RPC based servers >on the PC. (After all, we don't want to undercut our workstation >business :^) "rwall" uses a pmap_rmtcall to contact rpc.rwalld, but >to be correct you have to handle both direct and indirect calls, don't >you? We didn't implement the portmapper in total, we have a port 111 interrupt routine which checks for pmap_rmtcall to rwalld. The 179 bytes refers to all code (including portmapper check routine) which is resident and executed above the UDP level. > >Also, are you doing the any duplicate filtering? > No. > >I'm sure you know that none of the resident PC-NFS code is written in C... > No I didn't know that. - Carl Beame Beame@McMaster.CA