jmg@cernvax.UUCP (jmg) (12/08/87)
This is a bit of a flame, which I hope does not upset some people too much. I have tried sending the comments privately, but have had no reply. I got a Kinetics internal Ethernet interface for a Mac SE, plus the ethernet driver, test software and NCSA telnet version 1.12. In order to try out this software in a safe manner I created a mini- -Ethernet with the Mac and an Ethernet monitor. Am I glad that I did this! The test software, when run, tends to throw out a large number of broadcast packets in a very short space of time. Sometimes one can control the frequency, other times not. At least one test threw out about 200 broadcast packets in much less than one second. If I had been on the real CERN Ethernet then a few hundred users would have had to deal with these! FLAME ON When will people writing test software avoid the intensive use of broadcast packets? Multicast would be slightly better, but even then the software should establish the address of those other computers with which it can run a test, and then address them directly. FLAME OFF (for a while) I then tried to run NCSA telnet. This also started out with about 70 immediate broadcasts. These started out with a set of three types of broadcast: 1. arp with source ip address 0.0.0.255, looking for 0.0.0.127 2. something with type field 80f3 (what the hell is this?) 3. some other arp-type (type field 809b) with sender as 0.0.0.127 These three are repeated about 20 times at intervals of about 10 milliseconds (yes, milliseconds!). There are then a few more type 809b broadcasts at reasonable (a few hundred milliseconds!) intervals before telnet starts to arp for the real host that I asked for. FLAME ON Why does software often insist on repeating packets at very short intervals on vey reliable LANs (and have you seen the Sun lately!)? FLAME OFF Despite all the above, I waited for a quiet moment before connecting onto the real Ethernet. I then tried telnet to our Ultrix Vax. Immediate remark: keyboard in application mode does not work for us. I then thought to run the vt100 test program (which some of you might also have picked up off usenet). What a disaster: the emulation fails all over the place! Never mind, let us see if I can connect to our IBM VM system. Of course, I have to go via a Spartacus KNET, because there is no NCSA tn3270 (is anyone working on this?). Complete failure: Spartacus has a bit of a peculiar telnet setup (though Ultrix, bsd4.2 and FTP Inc. telnet on a PC work fine) which seems to screw NCSA telnet. Final try: go through an IBM 7171 front-end, which has 3270 to VT100 built in. Sort of works (using ESC n for PF key n), but since the application keypad mode fails there is no way that I could get PA2 for clear screen. Merde (which the French will understand. FLAME ON I know that NCSA is now at version 2.0. Why did I get version 1.12 from Kinetics? (and why must only a Kinetics agent modify their Mac SCSI box for a European power supply?). How do I get an updated version quickly (no, I cannot do anonymous FTP!). Why have these simple tests never been reported before? etc. etc. FLAME OFF I would be delighted if someone could tell me that all the above problems are fixed in the current release!
krauskpf@uxe.cso.uiuc.edu (12/18/87)
After this message was posted on usenet, Mike's message was posted with a reply on our telnet mailing list (telnet@ncsa.uiuc.edu). This one showed up on our notesfiles two weeks later, so here is the same answer. Anyway, the answers are: Yes, these things are generally fixed in version 2.1, except for the keyboard problem. We have been talking with Apple's technical people to see what the fix is. Anyone with the source code want to fix it and report back? With Kinetic's products, the FastPath, EtherPort SE and EtherSC, the gateway software provided makes use of too many broadcast packets, especially the Ether products. Please inform Kinetics, as we have, that this is unacceptable. The soon to be released EtherTalk support from Kinetics, used with telnet 2.1E (EtherTalk version), lets NCSA Telnet go directly to the Ethernet and cause no extra broadcasts. Kinetics is now shipping version 2.1 with their products. We don't rate our VT100 emulation on test programs, although we have used test programs to help debug our code. Do you have any applications which fail under NCSA Telnet's VT102 emulation? Would you care to report the exact bug or contribute a source code modification? For the keyboard, you can use the control panel to re-select your keyboard when you exit NCSA Telnet instead of re-booting. NCSA Telnet for the Macintosh (and PC) is available from: Anonymous FTP from ftp.ncsa.uiuc.edu (128.174.20.50). Tim Krauskopf National Center for Supercomputing Applications timk@ncsa.uiuc.edu