zlsiial@cs.man.ac.uk (04/18/91)
I am writing this in a state of mounting fury, though I doubt that will abort any of the flames. I manage a group of 11 HP 9000/300 clusters, about 90 processors, at the University of Manchester. We have a large number (about 900) of hosts attached to a local ethernet system, and we are connected to the rest of the world via a gateway to the JANET network. The hosts file on our systems is maintained by another department, and, since this is an active university, it changes frequently. The maintainers of this list feel, quite rightly, that they cannot enforce any policies of name format; it is hard enough ensuring that all IP addresses are unique. I certainly have no time to rewrite a hosts file every few weeks. What I mean to say is that the hosts file is, in practice, a given which I cannot mess about with. When we install an HP cluster, users of the new machines can send electronic mail quite happily to any other machine on the local network. Naturally e-mail is used fairly heavily here. But we would also like to be able to communicate with the rest of the world. More than 2 years ago HP promised to supply us with an implementation of the Coloured Book software, which would allow our machines to communicate with other JANET sites and, ultimately, with the rest of the world. Stopgap software has indeed appeared from time to time, but as yet it does not run without causing an entire cluster to panic and halt from time to time. The problem is becoming less frequent. But another problem has reared its head. When the Coloured Book software is installed, it rewrites the sendmail.cf file. It does this in such a way that 1) electronic mail to JANET hosts becomes possible; and 2) eletronic mail via SNMP to local hosts becomes impossible. This means that a working mail system ceases to work as soon as the software is installed. In practice we can allow mail to work to remote users, but not to local users, or we can allow mail to local users, but not to remote users. Since installing thes software causes working software to stop working, it is clear that there is a bug in the software. This problem has been reported twice now to HP. According to them, the problem arises because the names of machines in our hosts file do not conform to the presuppositions made by the authors of the software. Accordingly they have made no attempt whatsoever to suggest a practical way out of this difficulty, and they have refused to spend any time on serious attempts to solve it. An optimum solution would seem to be this: if a host is in /etc/hosts, then use SNMP for electronic mail. Otherwise use Coloured Book software. Of course, no one knows enough about the format of sendmail.cf to begin to understand what has been done to it, much less how to fix it. This seems as though it ought to be a problem with a simple solution. Unfortunately, between the uncooperativeness of Hewlett Packard's support personnel, and perhaps the unwillingness of the designers of this wretched software to admit that it is very poorly conceived, no solution seems to be in sight. 1) Is there anyone familiar with sendmail.cf and with the Coloured Book's mangling of this file who can suggest a change? 2) In this as in half a dozen other irritating instances, HP seem to be fond of taking no action unless they are prodded into doing so. For example, they seem to have mislaid a compiler upgrade which we were supposed to receive six months ago. Weekly telephone calls produce repeated assurances that something will be done eventually. How many of us are suffering from this kind of brush off? A. V. Le Blanc University of Manchester ZLSIIAL@uk.ac.mcc.cms
ccrth@lut.ac.uk (Rob Thirlby) (04/19/91)
Organization : Loughborough University, UK. Keywords: We run a coloured books/SMTP mixed sendmail system on our HPs by ignoring the HP supplied sendmail generating procedures. I can supply working sendmail.cf files for a "gateway" machine on janet and SMTP and a slave config for other systems including HP SUN Apollo etc. I wondser if what we really need is an Edinburgh University Coloured Books user group as Ed supply the software to most UK unix vendors and the config problems are universal. We are likely soon though to go over to a system with a PP based mail distributor on the JNT sun system with local delivery by means not quite decided!! Rob Thirlby, network manager, Computer centre, Loughborough University. e PS HPs support in the Uk is fine as long as you dont run coloured books or an ex Apollo system! Rth.
jim@cs.strath.ac.uk (Jim Reid) (04/19/91)
>> 1) Is there anyone familiar with sendmail.cf and with the >> Coloured Book's mangling of this file who can suggest >> a change? The answer is (drum roll and fanfare of trumpets) to throw away the vendor-supplied sendmail.cf files and get hold of UK-sendmail. This is a package that churns out sendmail.cf files. You tell it what mail delivery agents you have and what hosts are reachable through them and it goes away and creates a sendmail.cf file to handle mail in precisely that way. It can do all sorts of other clever things too (like hiding a bunch of nodes under a single domain and flipping the domain ordering). Another PD program (C-nrs) will spit out the JANET table files for UK-sendmail from the NRS DERFIL. The advantage of this approach are simple: your mail works properly and you never have to worry about sendmail.cf files. The down side is that HP won't support anything other than the near-useless sendmail.cf files they provide. However, as these don't do what you want and the alternative works, this is not a problem. UK-sendmail and C-nrs are available from various sources archives in the UK. To save me digging out the details, these are also now available for anonymous NIFTP from uk.ac.strath.cs: Filename: <GUEST>/c-nrs.t.Z or <GUEST>/uksendmail.t.Z Username: guest Password: your-email-address Both files are compressed tar archives, so be sure to use the binary option when transferring them. The packages come with plenty of documentation and examples of config files. Jim