Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cs.utexas.edu!ut-emx!ibmchs!auschs!awdprime!gumby!shepler From: shepler@gumby..austin.ibm.com ( Spencer Shepler ph:3-4368 off:802/3E-71 ) Newsgroups: comp.sys.ibm.pc.rt Subject: Re: curious Keywords: IBM, AIX, RT, Opinions. Message-ID: <708@awdprime.UUCP> Date: 13 Oct 89 18:48:38 GMT References: <27415@iuvax.cs.indiana.edu> <17120@rpp386.cactus.org> <899@gort.cs.utexas.edu> <178@ursa-major.SPDCC.COM> <808@cwjcc.CWRU.Edu> Sender: news@awdprime.UUCP Reply-To: @cs.utexas.edu!ibmaus!auschs!gumby.austin.ibm.com!shepler Organization: IBM AWD, Austin, TX Lines: 65 In article <808@cwjcc.CWRU.Edu> chet@kiwi.INS.CWRU.Edu (Chet Ramey) writes: >Jason Levitt asks: >>>Could we get some feedback from people who are not working at IBM > >and Steve Dyer obliges: >>I haven't had a whole lot of AIX/RT experience. My brief experience with >>led me to conclude that the latest (2.2.1) wasn't too bad, although it >>still gave you this weird feeling of coming out of an entirely different >>"culture", although it had now seemed to have completed "UNIX as a 2nd >>language." :-) > >It flunked the networking part of the course, though. We tried doing some >development on RTs (135s, 115s) running AIX 2.2.1 and were just stopped >dead in our tracks by the awful NFS implementation. Programs residing on >an NFS-mounted file system got memory faults and dumped core 60-70 % of >the times they were executed. Running dbx on a program residing on an >NFS-mounted file system could reproducibly crash the system. The TCP/IP >implementation was hard to install and configure and fragile when it was >finally running. The name server interface (gethostbyname() and so on) >could not properly reverse-map IP addresses to host names (ftp 129.22.8.15 >would sit there for a while, then report failure trying to do a name >lookup). This is after you figured out how to set the name server up in >the first place, which was explained nowhere. I could go on and on... > >Other than that, the things I missed were job control (of course :-), the >user interface of the BSD terminal driver (though I got to kind of like >the Sys 5 termio programming interface before we pitched AIX), and a lot >of little BSD utilities (like gprof, more, diff -c, stty). > I am currently part of the NFS development team here at IBM Austin and would like to address some of the comments that Chet had to make about the NFS implementation on an RT running AIX 2.2.1. FYI: I spoke with Chet this morning and tried to find out exactly what kind of configuration that was making NFS fail so badly. I wasn't of much help because they had alreadly moved to AOS on their RT's there at Case Western. We did discuss some of the problems however. I would like to mention that here in our development group we use the AIX NFS implementation exclusively for all of our remote compilations and executions. It is used heavily over ethernet and token ring without any problems except for the occasional NFS Server not responding. The problems that Chet had with dbx may have been caused by the fact that the current NFS implementation does not support memory mapped files. As far as the problems with such a high rate of execution problems over NFS, I wasn't able to address that directly. Chet had mentioned that there may have been a Bourne shell problem that might have been part of the problem. My first reaction to Chet's posting was of total shock. We had attended this year's Connectathon with very good results and, as I have mentioned, have very few problems here. I hope those of you giving AIX the once over will give our NFS implementation a longer look before passing judgement. I don't work for marketing and don't have a desire to. I am just a developer that is concern about his product's performance in the field. Thanks, Spencer Shepler IBM Advanced Workstations Div. 11400 Burnet Rd. Austin, Texas 78759-2502 (512)823-4368