NOS FAQ draft 15-Dec-92 [This is a draft -- forgive ugliness, and help repair it!] This paper discusses the TCP/IP implementation known as NOS, originated by Phil Karn and developed by many others. It tries not to duplicate material that is available elsewhere, but refers instead to existing documentation. If you have never seen NOS before, you might want to scan this document quickly, and then go read the "Introduction to NOS" file by John Ackermann or the "Beginner's Guide to TCP/IP" (both described later). If you maintain or use a NOS variant, please check its description here in this paper. If it's wrong, or if there isn't one, please tell the maintainer. Other additions and corrections are also welcome. This document is maintained by mg@bds.com, as of December 1992. ---------------------------------------------------------------------- Repositories The best place to get NOS is from someone you know who is already using it. You can get it over the Internet from several anonymous FTP sites. The best-known site is ucsd.edu. ucsd.edu:/hamradio/packet/tcpip grivel.une.edu.au (Australian mirror of ucsd.edu) [ ucsd needs a README with one-liners for each directory ] It is also found on many telephone BBS systems, including: N8EMR's Ham BBS (614) 895-2553 ChowdaNet (401) 331-0334 V.32 (Fidonet 1:323/120) ChowdaNet (401) 331-0907 V.32bis The Antenna Farm (512) 444-1052 V.32bis N81 WB3FFV (301) ? Gracilis BBS ? AMNET BBS +61-3-366-7055 (Australia) N1BEE distributes a cleaned-up version of the GRI code. The primary distribution point is ChowdaNet. This seems to be one of the more stable and solid end-user releases, though the last version is nearly a year old now. Documentation Unfortanately, NOS is poorly documented compared to many other programs. This is not due to a lack of effort by the developers. It is because there are so many versions of NOS, being developed by many people who have no regular contact, and each version changes very frequently. Any documentation quickly becomes out of date. For some features, there is no documentation at all. Other features that are documented aren't present in all executables, even those produced by the same source, because that feature may have been turned off in the executable you are using. Thus, it is likely that the documentation you can find won't exactly match the program you are using, and you will have to refer to a full "base" manual and to another document that describes the differences from the base version. intronos.zip This is the best introduction to NOS for people who understand something about packet radio. userman.zip A complete user manual for the last pre-NOS version of NET... 890421.1 nos_1229.man A complete user manual for PA0GRI NOS. help_v02.zip mailbox help files (for ka9q 901130) The several volumes of the ARRL Computer Networking Conference (CNC) Proceedings contain many papers on TCP/IP and NOS. These are available from ARRL Headquarters in Newington, CT. Send mail to info@arrl.org for an automatic response pointing at more information about the ARRL. Particularly useful CNC articles are listed in the bibliography at the end of this document. Some of these papers are available online in the directory ucsd.edu:/hamradio/packet/tcpip/docs. Other papers are also available on ucsd.edu: ka9qnos.ps.Z netnix-paper.ps.Z connex.Z 1987 Directory of papers from CNC 1987 1988 Ditto 1988 There is no document dedicated to describing the internals of NOS. There is a section of the user manual that describes some of the socket library calls. Look in the section on NOS Internals here in this document. John Ackermann, AG9V offers: I hesitate to make the offer, but on a time-permits basis I can provide a hard-copy set of my doc, the PA0GRI ref manual, the Rutgers tcp/ip intro, and a disk with N1BEE's GRINOS for the cost of copying and mailing -- usually $10 or $15. It's best to call, write, or email first to find out what the turnaround time is likely to be. My mailing address: John Ackermann AG9V 2371 Stewart Road Xenia, OH 45385 Gary Ford has written a "Beginner's Guide to TCP/IP on the Amateur Packet Radio Network Using the KA9Q Internet Software" alternately titled "End User's Guide to TCP/IP". Version 2.0 of the guide documents the use of NOS 911229 (PA0GRI v2.0h) and BM v3.3.2. It is available on: ftp.eecs.ucdavis.edu:pub/ka9q or ucsd.edu:hamradio/packet/tcpip/docs nosbgnlp.zip unzips to a 66 page ascii "line printer" document, which can be printed with the UNIX lpr command. nosbgnps.zip unzips to a 54 page PostScript document, which can be printed on a PostScript printer. nosbgnpr.zip unzips to a 66 page ascii "print" document, which can be printed using the DOS print command. You will need to send control codes to your printer to control the page offset and you should turn perforation skip off. (not on ucsd.edu) Mailing lists rec.radio.amateur.packet (Usenet newsgroup) This is the place to ask beginner questions, such as "How can I use a baycom modem with NOS?" tcp-group@ucsd.edu This is a forum for NOS developers to discuss things they're doing. It is NOT the right place to ask beginner questions. Send mail to listserv@ucsd.edu with the word HELP as its only contents. It will send you instructions by return mail. nos-bbs@hydra.carleton.ca The purpose of this list is discussion of the ins and outs of running KA9Q NOS as something approximating a full-service BBS, which generally boils down to discussion of the care and feeding of the JNOS version of NOS, maintained by Johan, WG7J. Discussion of peripheral issues which are likely to be of interest to NOS BBS sysops, such as the convers server, NNTP, POP, etc, are also welcome. Submissions to the list go to: nos-bbs@hydra.carleton.ca Subscription/deletion requests and other administrivia to: nos-bbs-request@hydra.carleton.ca Note that the reply path for list mail is set to go to the list address, not the originator. That's by design - to encourage *public* discussion. If you want to send private mail to a nos-bbs contributor, please take care to get the address right! Operating systems and other machines NOS has been ported to several operating systems. Most development still happens under DOS. At some point, all the different platforms could have been built from a single set of sources. Since then, they have diverged enough that re-integrating the changes back into a common set of sources would take some effort. writes: When the list of targets was really small, in about 1988-9 maybe, we had a single set of sources that built on various targets... the 890421.1 release supported DOS/Borland and a variety of Unix variants. I haven't paid much attention to this since then. OS/2 Windows Macintosh Amiga Installation and setup tools An installation program for some versions of NOS is available. The NOS_KIT package explains well the several files used by NOS for configuration. It is a package for installation of the 901130 versions of KA9Q and G1EMM NOS (it includes executables for 901130 G1EMM and KA9Q). This "kit" is a reasonable place for a newcomer to get started. Written: Dave Fritsche (wb8zxu) E-mail: dlf@phx.mcd.mot.com Version: 910324 Path: UCSD.EDU /hamradio/packet/tcpip/install/nos_kit.* Dave writes: The "kit" should be unpacked and placed onto floppy disks. It all fits on a single 720k or larger disk. This disk is then used as a NOS installation disk (the files can also just be stuffed into a subdirectory on a harddisk if you prefer -- but put them in a directory close to root (c:\) -- long pathnames screw things up). Once the installation disk is ready (I hand them out to locals), just put the disk in a drive, change to the drive (e.g: a:), then type "install". The user is presented with two screens of "fill in the blanks" kind of data. Just use the , , , , and to move between the entries. It's pretty self-explanatory. There are a couple of questions that weren't very clear, that should be cleaned up. But it's something to help get a person off on the right foot. After the blanks have been filled in, the installer goes off and makes all the needed subdirectories, creates an autoexec, ftpusers, domain.txt, . . ., and then unpacks the binaries and all the support files. The "kit_src.*" files, are the source code for the "install" program contained in the kit. Numerous people have recently asked me for the source code, so that they could update the installation kit for GRI/WG7J NOS. Sure hope they can pull it off! Wish I had the time to do it myself. Hopefully, Brian will get this source code moved into the "install" directory at some time in the near future. Generally speaking, endusers won't need or want these source archives. Versions The main NOS repository harbors a bewildering variety of NOS variants. All of them are derived in some way from some ancestor of the KA9Q version. The one that is best for you depends on what you want to use it for. One reason why there are so many versions of NOS is that it is used for widely different purposes by different groups. Each one needs it to solve a particular problem, or provide some service, so they implement whatever feature they need. There are many such groups throughout the world, and many of them have no regular contact with any of the others. Each NOS variant tends to reflect the cumulative efforts of a person or group, rather than a particular added feature. Some people modify more than one area -- for instance, they might add a new TCP server, and "enhance" the way datagrams are routed. The variants are usually known by the names of their authors, and they often have nicknames. Some of the more common ones are ka9q wg7j (jnos) pa0gri grinos based on PA0GRI, cleaned up by N1BEE pe1chl wnos hrlnos gpsnos Georgia Packet Switch, produced by GRAPES (Georgia Radio Amateur Packet Enthusiasts) gracilis wampes g1emm pmnos Most versions of NOS share a common core of commands. Most versions of NOS have all of the following facilities in them, but emphasize one over the others: packet switch services UI terminal For example, GPSNOS is optimized to be a standalone packet switch, and doesn't offer any other services. (This is what DOS NOS is really best at.) WNOS has a nice split-screen user interface, and PMNOS has an even more elaborate one. The features supported by NOS can be divided into categories this way: network NETROM, FlexNet user-interface split-window, fkey, command recall hardware special serial ports, network interfaces services callbook server, pop, smtp, convers, mailbox Here is a list of the features by category: Network TCPIP AX.25 NETROM WAMPES AX.25 routing WNOS, WAMPES hardware asy standard PC serial port hs high-speed serial driver for 8530 (no DMA) scc generic 8530 driver DRSI EAGLE PI PACKTWIN dialer services ttylink (chat) mbox bbs convers pop smtp nntp ftp finger callbook gracilis user-interface split-screen (WNOS) scrolling, cut/paste, mouse (PMNOS) fkey ---------------------------------------------------------------------- These versions have been optimized as packet switches: ka9q gpsnos pe1chl nos pa0gri gracilis These versions emphasize user-interface: hrlnos minihrl wnos pmnos These versions emphasize services: wg7j ---------------------------------------------------------------------- Here are brief descriptions of some NOS variants. If your favorite one isn't here, or if you can describe it better, tell me! I'm particularly interested in documenting ancestry where possible. net Phil Karn The last pre-NOS version of NET... 890421.1 documentation in userman.zip ka9q 21-Jun-92 Phil Karn This is the base version from which the others are derived. [ I haven't tried this one yet. - mg ] Latest version on ucsd is 920621. wb8zxu: A lot of people consider 901130 the last "really good" working version of NOS. After that version, NOS got real fat, and fairly quirky. It also split about 5 different ways . . . PA0GRI, WG7J, PMNOS, WNOS, . . . wg7j 15-Dec-92 Johan K. Reinalda WG7J Also called JNOS. exchanges mail with W0RLI-style bbs systems split-screen ttylink sessions command recall (as of 1.07) converse ip and tcp access control accesses all dos drives (not just "root" one) This is one of the more actively developing versions. Based on KA9Q 911229 release, at least up to JNOS 1.07. Later versions will be based on KA9Q 920618. This is a service-oriented NOS. It acts as a "full-service" bbs. It can exchange mail with conventional amateur BBS systems, such as MSYS and W0RLI, as well as via SMTP and NNTP. This version is also widely used as a gateway between internet and amateur packet radio networks. This has a rewritten 8250 UART driver. [I've found it to be slower than the previous one -- see later -mg] Derived somehow from was0206, pa0gri, wnos. pa0gri 29-Dec-91 Gerard van der Grinten, PA0GRI This is one of the more actively developing versions. This is used as the basis for other versions, including gracilis. nos_1229 gri20m mods allowing accessing many disk drives by FTP server made for GRINOS 1.8b, draft version for 2.0l and (in nearest future) 2.0m available by anonymous FTP on zfja-gate.fuw.edu.pl (name GRI20L-1.ZIP). Jerzy Tarasiuk N1BEE distributes a cleaned-up version of the GRI code; this cleaned up version is called GRINOS; it is distinct from pa0gri's version. grinos Mike Bilow N1BEE The primary distribution point is ChowdaNet. (GRINOS is not a synonym for PA0GRI NOS.) This is a cleaned-up and packaged version based on the pa0gri. It is one of the more stable and solid end-user releases. Mike tries to make sure that his bug fixes are recycled into the main PA0GRI releases, so you only tend to run a version or two ahead on bug fixes with GRINOS. gracilis info@gracilis.com Don Lemley N4PCR Milt Heath There seem to be several "gracilis" versions. One is publically available on ucsd.edu. It seems to be derived from pa0gri. Gracilis has a special version of NOS that runs in their PackeTen -- it doesn't run on a PC. The Gracilis PackeTen is a standalone packet switch with a special communications processor (the Motorola 68302). This standalone version is derived from work described in a paper on NOSINABOX by Bdale Garbee, Don Lemley, and Milt Heath in the CNC proceedings. There are rumors that Gracilis has also substantially reworked NOS. This version doesn't seem to be publically available in source form, but they supply it with their PackeTwin (and PackeTen) modem/radio hardware. [Are the binaries available, and will they support other hardware devices?] There is a mailing list that discusses Gracilis products. To subscribe, send mail to listserv@knuth.mtsu.edu with an empty subject line and a message body (beginning in Column 1) with the command: SUBSCRIBE GRACILIS-L _your_address_here_ Says bdale: NOS is integral, and doesn't look a whole lot like NOS internally any more. Don Lemley and Milt Heath at Gracilis acquired a commercial text-based window system with source, and heavily modified it to work in a mutli-task environment. They also found and have integrated an overlay manager to deal with the memory size problems. Don has spent considerable time finding and squashing memory leaks, and other problems in NOS. What they ended up with is a completely different user interface to a package that does all of what NOS does, and more. wnos 14-Jan-92 Michael Bentrup (DB3FL) Mike Chace (G6DHU @ GB7IMB) ucsd.edu:/hamradio/packet/tcpip/wnos A very detailed set of manuals for WNOS3 and 4 is in the wnos directory on ucsd.edu. The author of this software Michael Bentrup (DB3FL). He is sadly not connected to the Internet. Mike Chace (G6DHU @ GB7IMB) produces a version of the software more suited to the UK environment where, for example, NET/ROM is required. WNOS was the first system to port two important features of the WAMPES Unix software namely the AX.25 autorouting front-end and the convers server. As far as I know, ALL flavours of NOS that support the convers server have used the WNOS code and therefore should be compatible. A unique feature of WNOS is the AX.25 autorouter. The WNOS front end allows a system to network at Level 2. The WAMPES AX.25 front-end stores paths to other AX.25 systems and users may use these paths without reference to the full path. Each system along the path simply looks up the next hop, opens a connection at L2 and relays the frames. All this is transparent to any user, be they an ordinary L2 user or someone using L2 to push TCP/IP through a network. Example G6DHU -> G4WRW -> G7XXX -> G8YYY If G4WRW has a route to G8YYY (via G7XXX), G6DHU can use this route without reference to the full path. In other words, all G6DHU need do to open a connection to G8YYY is say connect G8YYY via G4WRW as soon as G4WRW sees the connect request, he will open a connection to G7YYY via G7XXX. Therefore, each link on the path is hop-to-hop acknowledged and the end users only need to know their nearest link in the path and the final destination in order to be routed through the system. Also, any traffic directed through your system in this way, will have the path information saved and therefore available for future use. WNOS is also unique in that it remembers (and saves to disk) ALL of the routing table information AX.25 paths (as discussed above) NET/ROM routing table ARP table IP routing table Therefore, with WNOS it is quite feasible to start with a system completely devoid of routing table information and then get to know all this information dynamically. Also, as paths etc change, tables are automatically saved and updated on disk. It is a neat system! WNOS was also the first version of NOS to support real time data compression using LZW coding. email (via SMTP), news articles (via NNTP) and convers interlink traffic may all be transferred in data compression mode. Data is compressed before sending and results in an average of 45-55% reduction in data transferred against plain-text transfer. Another item of interest is the NNTP system. WNOS contains everything that allows a user to deal with news articles. There is a news poster and a news reader. The NNTP filesystem is automatically updated, for example, if an article arrives for a newsgroup unknown to the local system. The news spool is updated immediately and the articles accepted. WNOS also contains a special server which allows AX.25 mail to be converted to an NNTP news article and cross-posted to a newsgroup. This is most suited as an end-user node, rather than as a network service provider -- that is, it's a better terminal than bbs. Has a split-window user interface. A status line at the top of the screen displays info about the current window, and indicates when there is new activity on other windows. (Each "window" is really a full screen. You can switch among screens, but they don't overlap.) There is an input area at the bottom of the screen. attribute command sets monitor type . Supports convers, though probably not compatible with JNOS. The 'w' in the name is for WAMPES, which is an AX.25 routing mechanism (used in FlexNet?). By the way, I had a quick look in WNOS docs and it seems WNOS is ONLY a packet switch, it supports neither SMTP nor BBS message commands. hrlnos 19-Mar-92 by R. Kolb PA3EUG pa3eug@pi8hrl.ampr.org; PA3EUG @ ON4UBO; Derived from PA0GRI 16-Aug-91. Has a split-window user interface like WNOS. FlexNet: hop-to-hop-ack digipeating mbox log mem efficient mem thresh mode ipcam uses AX25 PID Text rather than IP. Does AX.25 autorouting. timers reworked. IP times are now dynamic. Saves arp cache to disk. minihrl is a smaller version of hrlnos. It has no netrom support, thereby gaining 80k of memory. gpsnos 11-Nov-91 Based on KA9Q NOS 910420. Optimized as a packet switch by GRAPES, B. Nebergall, K4TQL. Adds netrom/ax25 switch optimizations, removes mailbox functions. Has a remote sysop facility with a good "public" password protection scheme. Supports up to 5 hardware ports, including 56K ports. pmnos Presentation manager NOS, Walt Corey KZ1F The authoritative drop site for PMNOS is UCSD.EDU. The current beta version of PMNOS is PMNOS 1c: PMNOS1C.ZIP THIS IS STILL BETA! Please don't distribute this code without making sure that it is accompanied by: PMAIL.ZIP 9/18/92 -- 14:40:04 pmreadme.1st 10/6/92 -- (update includes this message) Users should have access to nos_1229.man(lp) (GRINOS manual) and the readme files from JNOS 1.04... There is NO OTHER documentation YET.... The "cleaned&cleared" release will be posted on UCSD.EDU and the source code will be integrated with JNOS. After Johan releases JNOS 1.05 (soon hopefully) a "current working" compile of PMNOS will appear at UCSD.EDU and at WG7J.AMPR.ORG.... The OS/2 and PM sources will be combined with the rest of JNOS and an OS/2 compile selected by #ifdef statements... IF everything goes well in the project. Currently, The IBM C set/2 compiler is the only compiler that has been used to build PMNOS. I have experimented with the GNU C/C++ compiler, but have had no luck so far (I am not a "real programmer"). Other compilers such as the forthcoming Borland, Zortech, and Watcom compilers for OS/2 should work in competent hands.... But these are still unknowns. WARNING-- if you have not programmed the PM environment, you had better go "larval stage" with the OS/2 'redbooks' (and other PMwindow specific DOCS!!) for a while before getting creative.... Stick to customized builds of the code for YOUR OWN PURPOSES. PLEASE don't distribute new compiles of this stuff!!! This ain't just DOS anymore!! The 'redbooks' can be downloaded from software.watson.ibm.com by anonymous FTP. Authoritative answers to questions on communication problems for OS/2 version 2.0 can be found on HOBBES.NMSU.EDU and SOFTWARE.WATSON.IBM.COM. Network driver support should be along later in the year: Planned are SCC, PI, and an NDIS 'shim'. The SCC and PI drivers will be loaded at IPL (in the config.sys schema) as: device=C:\OS2\SCC.SYS The NDIS 'shim' will enable use of NDIS drivers in a manner similar to the NDIS_PKT packet driver set. You will have to scrounge up your own copy of the appropriate OS/2 NDIS driver, protman.sys, etc..... All of the above is subject to negation, reversal, alteration, etc., by Walt Corey. wampes 16-Sep-92 Dieter DK5SG / N0PRA ucsd.edu:/hamradio/packet/tcpip/wampes WAMPES is a Unix based system derived from KA9Q. It supports all the usual TCP, UDP and IP services and was the first to provide the convers system (like the Internet's IRC). It was written by Dieter Deyke (DK5SG) and ran exclusively on machines running the HP-UX Unix flavour. It has since been ported to ISC and SCO Unix running on PCs and within the last few monts it has been sucessfully ported to Linux, the free Unix for 3/486 PCs. It currently supports HP 9000 series 300, 400, 700, and 800 computers, running HP-UX 08.xx. It at least compiles on SunOS, but hasn't been thoroughly tested. Alan Cox On the subject of wampes there is also a version for interactive unix and allegedly one for linux (tho I've not traced this). [How is this related to HRLNOS? ] k5jb Joe Buswell, K5JB, Midwest City OK Packet Address: k5jb@k5jb.ok.usa.na Amateur Radio IP Address: 44.78.0.2 Internet: jbus@sabea-oc.af.mil CIS: 70305,1341 This is a variant of ka9q (890421) that runs on SCO Unix. pe1chl 1-Jul-92 Network netrom fixes "ftl0" and "broadcast" protocols/command sets for access to the microsats like AO-16 and UO-22 can receive (but not transmit) the PACSAT broadcast protocol, which is sent by the PACSAT amateur satellites. UI changes fkey version screen services bbs forwarding scripts Costas SV1XV (ex G7AHN) Regarding MINIHRL, the author PA3EUG states it is HRLNOS without the NETROM code. I have been unable too to find the full HRL NOS code, it is not on UCSD.EDU or on any other well known site. PE1CHL is a release of NET.EXE with very limited user services very simple mailbox but it is reasonably well designed for ax25 and also supports the "ftl0" and "broadcast" protocols/command sets for access to the microsats like AO-16 and UO-22. We might run it here in Athnes on SV1IW's station to create a TCP/IP sat gateway. Good docs on PE1CHL are files NETDOC1.ARC and NETDOC2.ARC in directroy /hamradio/packet/tcpip/pe1chl on ucsd.edu G0BSX mail server: ucsd.edu:/hamradio/packet/tcpip/g0bsx/server.tar.Z g1emm Kelvin Hill G1EMM ucsd.edu:/hamradio/packet/tcpip/g1emm This was once a very actively developed version, but doesn't seem to have grown much in the past year or two. It branched off into PA0GRI (and hence GRINOS). SV1XV, Costas writes: Kelvin also distributes SMALLEMM.EXE (without AX.25/NETROM/KISS, which only supports SLIP and Ethernet interfaces) and is very small. I use it on my laptop Amstrad PPC-640D. He also distributes an example of a full G1EMM NOS installation in the file G1EMMKIT.ZIP. N8GNJ, Steve Stroh writes: I think that when Kelvin got out of NOS codesmithing, PA0GRI took his code and started the now popular PA0GRI variant of NOS. The G1EMM code is effectively obsolete, but it is probably maintained on several systems. unknown Alan Cox writes: There will be a ka9q net with smtp links to the O/S, host mode login via ax25 and the ability to act like a netrom node for Linux very soon. I'm just finishing debugging it. If I get it finished I'll mail you if you want to stick it on the list. ---------------------------------------------------------------------- NOS Internals There is not a whole lot of documentation on NOS internals. The socket interface used by the TCP clients and servers closely resembles the one found in BSD Unix. The tricky parts arise in dealing with the multi-tasking kernel. More details on the NOS internals may be found in the following files: [What internals documentation is there?] Source code organization The sources are distributed in a single flat directory. The names of the files do not have prefixes that would identify the part of NOS to which they belong. The modules are divided into groups. Each group of modules is placed in a library. Most variants of NOS are distributed in two or more packages. There are no guidelines for what to put in a distribution, or how to package them, so each NOS variant is distributed differently. NOS is generally packaged appropriately for the system that runs it. DOS versions of NOS are almost always packaged in zip files. Unix versions appear as compressed tar archives, and less frequently as zip files. One package typically contains an executable, and possibly some sample init files. Another package contains the documentation specific to that variant. Another package often contains the full set of sources that was used to build the executable. It might seem silly to distribute the entire set of sources even when only a couple of things have changed. It is not practical to distribute packages of only those modules that have changed, because there is no standard base version to serve as a reference. Even if there were, that reference version would itself be changing, so people would have to keep several copies of it on hand. ---------------------------------------------------------------------- Running different NOS versions with the same configuration It is sometimes possible to use the same set of configuration files and tree of directories with several versions of NOS. Some versions are so different, however, that they will not be able to understand each other's config files. For instance, WNOS has trouble with the config files for JNOS. The mailbox help files (in spool/help) are often quite specific to a particular version. Even worse, the commands that a NOS understands depends on what features were turned on when it was compiled. If your config file contains commands to set up the NETROM interface, they will produce error messages if you run a NOS that was compiled without NETROM support turned on. There isn't really any way to fix this problem, and it isn't all that unreasonable. The following are some known inconsistencies that could be fixed by someone who has the time to do it. [Add to this list, please! Try to be as specific as possible. The less time someone has to spend fixing something, the more likely it will be fixed.] - The ftpusers file requires its fields to be separated by spaces in most versions of NOS. You can't use tabs. [This has been fixed in JNOS 1.07.] - The 'domain suffix' string must NOT have a leading period, but can (must?) have a trailing period. If you do it wrong, some versions will complain, but most will just quietly fail to work properly. I think it would be reasonable for the command to munge the string to look the way it should. - In most versions of NOS, the arguments to some commands in autoexec.nos MUST be separated from each other by SPACES. If you mix tabs and spaces, the command will be parsed incorrectly due to a bug in the parser. [This has been fixed in JNOS 1.07.] - There are minor variations in similar subcommands: ax25 retries vs. netrom retry I'd vote for the plural form myself, because the singular form might suggest that it is a boolean that controls whether a retry should happen, whereas it is actually an integer count. - The help strings that the commands print when given no arguments, as well as documentation, use "label" and "interface", "host", "address", and "destination" ambiguously. "address" could be an IP address, an AX.25 callsign, or an Ethernet address. "label" usually turns out to be the name of an interface. Ian Wade G3NRW has suggested the following conventions in his paper in the 10th (1991) ARRL CNC proceedings. an AX.25 MYCALL callsign (e.g. G3NRW-5) a host name in domain.txt (e.g. g3nrw or g3nrw.ampr.org.) an internet address (e.g. 44.131.5.2) or a user at a computer (e.g. ian) a device interface name (e.g. ax0) a device I/O base address (e.g., 0x3f8) an IRQ level (e.g. 4) [why not ? - mg] - Different commands do different things when invoked with no arguments. This in itself isn't so bad, except that you can't always tell whether doing so had an effect or not. Some commands print a help message when invoked with no arguments. Unfortunately, some simply print "needs at least one argument", instead of just telling what the choices are. (If you type just "ax25" it says that, if you type "ax25 ?" it shows you the subcommands.) Other commands that can take arguments display some status information when they are invoked with no arguments. Unfortunately, some of them don't appear to do anything -- you just get another prompt. This could happen if they would report some string and that string is empty, such as the "motd" command. Such commands should really print something like "", instead of appearing not to do anything. Moreover, if a command that could take arguments *does* do something when invoked with no arguments, it should say that it has done it. For instance, a command might set a variable a default value if invoked with the value omitted. (No commands that I know of do this; omitting a value causes them to print the current value.) Note that commands can't tell if they are being invoked from autoexec.nos or by someone typing interactively. This is probably why most of them aren't very verbose. ---------------------------------------------------------------------- Packet drivers The packet drivers from Clarkson and others are used to allow NOS to interface to many pieces of hardware and software. NOS can be interfaced to the G8BPQ switch and to BAYCOM modesms. Why are there still compiled-in device drivers? Why aren't they implemented as packet drivers? William Allen Simpson writes: Part of the reason is that packet drivers don't separate the encapsulation (link-layer) from the actual device (hardware-layer) very well. So, for example, when you have SLIP, PPP or NRS over two async ports, you would need to load all of the link-layer code twice. Also, a link-layer like PPP requires significant resources such as buffers and timers, which either need to be pre-allocated and managed for every device (wasting lots of CPU and memory), or managed under a single mechanism. BTW, Merit has built a PPP packet driver for NCSA Telnet. It's over 100K, instead of the 24K we have inside NOS. Therefore, I believe that KA9Q (Phil Karn) decided on this architecture to save CPU and memory. Remember, he started the program under CP/M! Now, there was some talk a year or two ago about a re-designed device driver mechanism to replace packet drivers. But there was also talk about going to NDIS drivers. Which shall we do? I can't seem to attach more that 3 packet drivers at once in NOS. Mike Bilow writes: That's true, you cannot attach more than 3 packet drivers. This would have to be changed in the source code. In fact, someone should probably finally sit down and make it correctly configurable with a simple compile time #define somewhere. ---------------------------------------------------------------------- Common problems Async interface loses characters If your machine can't field the interrupts from the serial port fast enough, it loses characters completely. One symptom is that datagrams get retransmitted frequently over a SLIP or PPP line. In some NOSes this causes the TCP connections to reset automatically after getting a few datagrams across. (TCP performs very poorly when the IP level datagrams are not delivered reliably, so it's best to just give up rather than waste bandwidth.) Another symptom is that garbled callsigns appear in the 'ax heard' list. One solution is simply to use a slower speed, like 4800 baud. A better solution is to replace the 8250 or 16450 UART with a 16550 UART that has a built-in character FIFO. This chip can buffer up to 16 characters while the machine is busy doing other things, so it relaxes the demand for prompt service. I've found that the JNOS i8250.c driver is slower than the others. It loses characters badly on a 9600 baud telephone modem, whereas other versions can keep up with no character losses at all. Rumor has it that DesqView may also cause problems if you run NOS in window 1. Serial communications-intensive applications should ideally be run in window 2, as that one is actually optimized for serial comm. John Ackermann, AG9V, writes: I'm not sure why Johan's 8250 driver is slower -- I personally haven't noticed that -- but I do know that at least through his 1.04 release the 16550 fifo was enabled; the code would detect the '550 but wouldn't do anything with it. In JNOS 1.05, the fifo size can be specified as an attach command parameter. Why do large amounts of core suddenly disappear? For instance, I started with 110K core, and FTP'ed a 1.5 Megabyte file off a LAN host (14K bytes/sec transfer). Core dropped to 30K. (using PA0GRI 2.0l) John Ackermann AG9V writes: We found that the disappearing core was due in very large part to the number and size of Ibufs you've configured. What happens (I'm not a Real Programmer, so this may not be technically right on) is that when NOS needs another ibuf, it has to find a big enough unfragmented chunk in the heap to hold it. If it can't find a large enough piece of memory, it goes to the core to get one. When you have ethernet-sized ibufs, it's quite likely that there won't be an unfragmented chunk big enough, so core is going to go away quickly. When we ran 1600 byte Ibufs, even starting with 190k coreleft, we could only run for about three days max before core would disappear. We reduced the ibufsize to 448, which despite manual references to the contrary seems to be the minimum that works with an MTU of 256 and now we see crashes resulting from loss of core (there are OTHER kinds of crashes, though...). So, sacrifice some performance on the ethernet by reducing your ethernet mtu, and your mss, to match what you'll really be running on the air. Then, set ibufsize to about that mtu plus 128. I haven't yet been able to determine whether the number of ibufs you set has any noticeable impact on this situation. That's one for the coders to answer. "Carl Makin" writes: Just one note. The Ibufsize must be larger than your largest buffer size. (Not the same as but larger otherwise the SCC driver doesn't work properly.) So I assume you are running attach buffersizes just a little bigger or the same size as the mtu of the interface. ie 447 mtu for an ethernet port with a buffersize of 447. John Ackermann AG9V writes: Actually, I've been running attach buffers much larger than ibufsize; never knew that was a requirement. And, it doesn't seem to affect things negatively. When I ran big ibufs, they were bigger than the attach buffers and when I had the memory leak problems. It would be nice for One Who Truly Knows to fully document these buffer issues... their tuning is critical, but there sure isn't much to go on. Witness the statement in the asystat command section of the reference manual (that's been there for ages) that the number of software high-water-mark hits tells you whether you need to adjust the buffer size, and then refers you to the attach command chapter for further details, of which there are none. Johan K. Reinalda, wg7j writes: Asy attach buffer doesn't care what the ibuf size is; they are fifo type buffers used during the interrupts, from which the asy rx process reads characters into mbufs. An asy buffer is allocated (with ints on) during the attach command, and is there for the life of the interface. Scc, packet and pi interfaces use the interrupt buffer scheme very actively for receiving of packets. They allocate buffers at interrupt time, and the queue need. The size of the buffer is somewhat complex. It needs to be the largest of -the largest ax25 paclen + about 100 (ax.25 header, actually 72 bytes max) -the largest ip mtu (if in vc mode over ax.25) + 20 (ip header) + about 100 -the ethernet mtu + safety margin :) I have some written up in the latest draft of the jnos40 DE manual, I will include that in the docs for 1.05. Why won't my PK-88 (and PCB-88) send/receive a packet larger than 1048 bytes? The AX.25 spec limits packets to 256 data bytes, but we'll ignore that. It's a built-in limitation of AEA's. Some Kiss implementations have even more severe limitations. The Kantronics ROMS (at least the earlier ones) won't allow 1024. You can try getting newer ROMS for your TNC from the manufacturer, but they can legitimately claim that they conform to AX.25 (KISS doesn't really specify the max packet length; AX.25 does). The TNC2 TAPR ROMS do not seem to impose any limitation other than memory. Why do I see what look like lower-case o characters scattered throughout the output of various NOS commands, such as the output of the /who command in converse? The "lower-case o" characters are TABs (ascii 9). You need to use the ansi.sys DOS screen driver (or one of its variants), which can properly expand the TABs. Place 'device=c:\dos\ansi.sys' somewhere in your config.sys file (using the correct path, of course). A popular ansi.sys variant is nansi.sys, available in the directory ucsd.edu:/hamradio/packet/tcpip/nansi. --------------------------------------------------------------------- Compiling NOS What compiler do I need to use? Most versions of NOS were developed using Borland (Turbo) C. You'll have the best luck with Borland C++ V3.1. Earlier versions may work, but beware of invisible compiler glitches (see next section). Originally, Borland sold the compiler, assembler, debugger, and profiler as "Turbo C Professional". Later they split the package into pieces. Now the compiler is available for less than $100 as "Turbo C++". The compiler and all the tools is available as "Borland C++", currently V3.1, and costs more than $300. (For another $200, you can also get Borland's "Application Framework" for building Windows applications, but that isn't needed for NOS.) Compiler glitches Turbo C++ 1.0 cannot compile NOS, because the compiler is too buggy. (It compiles it, but the resulting executable doesn't work right.) I believe that Turbo C++ 2.0 will successfully compile it, but you need to apply to the compiler patches supplied by Borland. Borland C++ V3.1 seems to produce reliable NOS binaries. BC++ V3.0 will work too, but people have reported that it produces an executable that crashes after about a half-hour (JNOS 1.07). Mike Bilow, writes: The BC++ 3.1 "-3" compiler switch is not safe with NOS. This appears to be limited to certain specific modules, primarily ALLOC.C, but I have never managed to encapsulate the problem sufficiently. What I think is happening is that the use of the "HUGE" keyword in some places causes a conflict with the "-3" switch. Compiling ALLOC.C to assembly source with the "-3" switch will show some very funny results, for example. On the other hand, the "-Z" compiler switch, which used to be unsafe, is now safe under BC++ 3.1. You win some, you lose some. You will find, I think, that Borland's "-O2" switch (optimize for speed) is generally not a good idea. I have found "-O1" to be stable as of BC++ 3.1, where it was not in prior versions. If you need faster code, you would be better off trying to force inlining of particular modules, especially inportb() and outportb(), which stand in the way of fast I/O if you have to put up with function call overhead on each use. Compiler warnings Most versions of NOS have code that the compiler considers questionable, so it will produce a few warning messages. Most of these are harmless, and are caused by functions not being declared properly. These warnings will probably be corrected later, but they don't cause real problems so they are not high on the authors' lists. If you only get warnings such as the following, don't worry. (These are from a compilation JNOS 1.07 with BCC 3.1.) Warning config.c 317: Suspicious pointer conversion Warning smtpserv.c 447: Function should return a value in function getmsgtxt Warning convers.c 911: Constant out of range in comparison in function channel_command Warning converse.c 1115: Constant out of range in comparison in function name_command Warning forward.c 466: Unreachable code in function makecl Warning tcpcmd.c 473: Suspicious pointer conversion in function doview Warning ipcmd.c 360: Suspicious pointer conversion in function doroute Warning arpcmd.c 342: Suspicious pointer conversion in function dumparp Warning arp.c 119: Condition is always false in function arp_input Warning arp.c 384: Possibly incorrect assignment in function arp_timeout Warning timer.c 61: Possibly incorrect assignment in function timerproc Warning alloc.c 405: Nonportable pointer comparison in function dofreelist Warning nrcmd.c 221: Suspicious pointer conversion in function doroutedump Warning nrcmd.c 328: Suspicious pointer conversion in function doallinfo Warning nr4.c 719: Suspicious pointer conversion in function nr4state What does 'linker fixup overflow' mean? Mike Bilow, writes: >Fixup overflow at _TEXT:0092 , target=__MKNAME >c:\bc\lib\cl.lib in modeule FCLOSE. You need to force all of the MKNAME.C module to be in the _TEXT segment. This is explained in the voluminous comment I put into MKNAME.C. You need to make sure that MAKEFILE invokes the compiler with the -zC_TEXT switch when compiling that specific module. What is happening is that library routine fclose() is making a call to __MKNAME(), which is declared to be "near pascal." This means that the linker must be able to resolve an address for __MKNAME() within 64K; ts is a "linker fixup." If __MKNAME() is more than 64K away from the place in flclose() where it is trying to call __MKNAME(), that is a "linker fixup overflow." This is not, by the way, a problem that can be finessed. It must be fixed. What do I do when DGROUP exceeds 64K? The error happens because too many data objects (variables, strings, etc.) have been placed in the data segment, which is restricted to be only 64k bytes long. The easiest way to avoid this is to choose fewer options in config.h, so fewer data objects are defined. However, setting -Ff=511 or -Ff=255 in the makefile this will usually do the trick, too. Another way is suggested by Mike Bilow : The DFAR kludge that I developed to finally fix the linker error about DGROUP exceeding 64K found its way into the distribution release of PA0GRI NOS as of v2.0l or so. Although it will work with Borland C++ from version 2.0 or later (and this is correctly handled by a #define in GLOBAL.H), the compiler must be invoked with the "-Ff" switch from the MAKEFILE. (Variants with the threshold specified, such as "-Ff=511", will work fine.) I have not looked at Johan's [wg7j] source lately and I don't know if he included the DFAR kludge also, but it is generally more stable to use the DFAR kludge where the programmer picks what gets expelled into the far data segments rather than the Borland "-Ff-nnn" automatic far data threshold. In my experience, "-Ff=nnn" will work with only sufficiently large values for "nnn," and will fail if smaller than 255. In other words, it works better to tell the compiler to be on the lookout for far data conversion with the "-Ff" switch, but to manually specify what gets moved using the DFAR kludge. The "-Ff" switch with no value specified defaults to 32K, which doesn't move anything automatically. (Believe it or not, no one has yet put a static data object in NOS that is bigger than 32K.) Where can I get the pklite program that the NOS makefile uses? pklite is in ucsd:/hamradio/packet/tcpip/util/pklte105.zip, and many other places. It's a companion program to pkzip et al. Can NOS be made to use code overlays, so only the parts that are really in use reside in memory? Mike Bilow, writes: Making NOS work with code overlays is a discussion that comes up from time to time. I spent a long time trying to get it to work, and all I ever had was a program that crashed almost instantly. The problem is that NOS changes the stack segment register on the fly, and the overlay manager has to know certain things about the stack in order to work. The general consensus is that overlays cannot be done, although I have never conceded that and can't provide a counterexample. ---------------------------------------------------------------------- DesqView, windows Mike Bilow, writes: You have some immediate problem when trying to port NOS to DesqView or Windows. First, Windows is somewhat unsuitable, since Windows apps expect to cooperatively multitask, and real-time performance is a thing requiring lots of kludges, even by Windows standards. There are direct solution for doing this kind of Windows programming, most obviously writing a custom VxD virtual device driver, but you may find that you have rewritten a good chunk of Unix by the time this works. DV is more suitable, but still leads to problems. It can be run on a very simple machine, although you don't gain much with DV until you have a 386 or better. At this point, you are probably looking at the whole difference between a good DV setup and a good OS/2 setup being 4 MB of RAM, about $120 at current U.S. street prices. I think DV is a really well done, even briliiant, product that fit a market niche that will soon disappear. Walt Corey [44.104.0.23] writes: As I have mentioned before windows is not a multitasking system, it is cooperatively multitasking. This means that only one program can run at a time (period) That one program can give up control to Windows at selected / predetermined points, at that point Windows can give control to another app. That program in turn runs until it decides to give up control. If for some reason (error perhaps) it doesnt, nothing else (including windows) will run. If a program "hogs" control for say 1 sec, at 9600 baud thats abt 1k data lost, for 5 sec its about 4.5k lost. It essentially could be done but there would be a huge risk of data loss and each process would have to be a window, as in the "dispatchable" unit of execution under Windows. Under OS/2 (for instance) it is a task not a window and OS/2 is a pure multitasking operating system, the timer task can not be stopped by a bad app, neither can the asy devices or the keyboard etc etc. chrisc@london-pride.lmt.mn.org writes: There is another dos multitasker that would seem to fit the bill here it's Free, well documented, source available (also free), and has almost all the facilities needed built in. The program is called CTASK written by (I think) T. Wagner in Germany. I had considered a while back cutting the socket lib out of NOS and moving it to CTASK but the job looked a lot bigger that the time I had available. Why does NOS implement its own task-switcher, when there are several that exist and do a passable job? DOS NOS was done before DesqView was viable. Also, requiring that someone have DesqView in order to run NOS would have limited the number of people who could run NOS. ---------------------------------------------------------------------- Questions [ This section should summarize the topics that arise perennially on tcp-group. ] How can I use NOS between computers over an RS232 wire link? NOS can use an asynchronous RS232 link by using a driver known as the async serial driver. Look for the documentation under the 'attach' command, using the 'asy' driver. The link protocol uses the async driver. Whereas the async driver will move the characters from one machine to another, the link protocol interprets them. Most versions of NOS support one of two link protocols. SLIP is the most commonly used serial port link protocol. PPP is an alternative. It is similar to SLIP, but permits the hosts to negotiate things like the interface's MTU automatically. If you don't know what that means, don't worry. You're probably better of using SLIP. Which of the RS232 signals are required on an async serial connection (i.e., the minimum required)? RXD, TXD, and ground (pins 2, 3, and 7) are the minimum required. If both ports expect to be connected to a modem (most PC serial ports do), you'll have to use a null modem to flip pins 2 and 3 (also 4 and 5; see below) between the ports. Beware that there is no hardware flow control with only these three signals. Unless the link protocol does software flow control [I doubt that SLIP or PPP do], characters may be lost. If you also connect RTS and CTS (pins 4 and 5), NOS can use them for hardware flow control. Does the async serial driver in NOS/NET do any hardware handshaking for flow control? Yes. There is more than one implementation of the async serial driver, and each one handles handshaking differently. The most widespread one is the one found in KA9Q NOS. With it, you specify whether it should use RTS/CTS handshaking by putting a 'c' in the options word at the end of the attach command [before the IP address?]. Another async driver, found in WG7J's JNOS [what others? GRI?], also supports RTS/CTS handshaking, but you don't have to specify that it should do so. It automatically detects whether the other side is using flow control, and acts appropriately. If you do specify a 'c' in the 'attach' command's options word, the driver will ignore it, so it doesn't hurt to leave it there. [ Then how can I tell it to initiate the handshaking? ] This async driver can also use the RLSD (also called DCD) signal for flow control. How do I use NOS as a router (gateway) between an Ethernet and an Amprnet via a TNC? Jack Spitznagel writes: You may use any ethernet card that you have a Russ Nelson/Clarkson/FTP Inc. Packet driver for. I have used NE1000/2000, DEPCA, 3COM, and WD/SMC8003/8013. Drivers also exist for IBM Token Ring and Arcnet adapters. As in most TCP/IP implementations the "ifconfig " command allows one to assign independent (or the same) ip addresses for each interface. Gatewaying is supported by *either* doing an "arp ... publish" for the remote slip ip-address at the connected ethernet end, *or* using the "ifconfig encap" to encapsulate one subnet thru another ones ip address link. (this is the amprnet gateway system you have heard talked about.) My explanation presupposes you are familiar with most of the TCP/IP routing and addressing issues. My machine crashes when I run it as an ethernet <-> slip router. Try getting a newer version of NOS (especially the KA9Q version). Also try running with handshaking totally disabled at both ends. Why do I see garbled callsigns in the 'ax heard' list? This can happen because of errors in the received packet. Bad packets occasionally occur in any transmission medium, especially on a shared radio channel. If the sending stations TxDelay is too short, you may also see garbled packets. Not all bad packets will appear in the various error counts kept by NOS. Garbled packets may also result if your serial port loses characters; see the section on 'async interface loses characters' elsewhere in this file. How do I get a permanent IP address? Talk to the AMPRNET address coordinator for your area. A list of them is available in ucsd.edu:/hamradio/ampr_coordinators [?] The best way to find out, though, is to find other people nearby who are already running an AMPRNET host. How do I set up a machine as a mail gateway? [this needs work - there MUST be a better way to do all this!] Briefly, add a REWRITE RULE to spool\rewrite FOR EACH HOST ALIAS. The machine must have a mail host address, one for each mail network that you want it to connect. (A "mail network" is simply a bunch of machines that can, directly or indirectly, send mail to each other.) The machine considers itself to have only one real hostname, which is set (in AUTOEXEC.NOS) via the 'hostname' [?] command. We'll call the other names "hostname aliases". With NOS, you must take care that your gateway machine recognizes that the mail is destined for it, regardless of which of its host names (or aliases) the message was sent to. To begin with, the machine only recognizes mail addressed to its own IP hostname (the one that's set by the NOS 'hostname' [?] command). To make it recognize its other host names, you have to add a rewrite rule in the file spool\rewrite that turns that hostname alias into the real hostname. If you fail to make your machine recognize all its hostname aliases, then when a message arrives for one of its aliases, it will forward the message to some other machine (usually to its SMTP gateway). Some machine down the line will realize that the message is supposed to go to your host, and will send it back there, creating a loop. Eventually, some machine in the loop will notice that the message has been to the same hosts several times, and will (hopefully) return it to the sender. What is "the mailer"? NOS can act as a maildrop, collecting and storing messages addressed to you. There are several ways it can do this, such as SMTP and "conventional" amateur BBS mail forwarding protocols. Also, people to connect or telnet to your mailbox can issue mailbox commands to leave you mail. No matter how the mail gets there, it is stored in files in the spool/mail directory, where it stays until you do something with it. All messages addressed to a particular recipient are concatenated together in a single file. The mailer is a program, separate from NOS, that shows you the messages in a mail file. If you are running plain DOS, you have to exit NOS to run the mailer. (You can also use the 'shell' command to suspend NOS and get to a DOS command interpreter, but that takes a lot of memory to work.) Then you run the mailer program to read your messages. You can delete them, reply to them, forward them, and so forth. When you're done, it updates your mail file, and queues any messages that you asked it to send. NOS will try to send the messages the next time you run it. You can run the mailer while NOS is running if you're running a multitasker like DesqView or DoubleDos. You can choose which program to use as the mailer by setting the DOS environment variable MAILER in your AUTOEXEC.BAT, as in SET MAILER=VIEW.EXE Some of the mailers that are available are: BM The earliest mailer that was included with NET/NOS. Written by Bdale Garbee, N3EUA. ucsd:/hamradio/packet/tcpip/bm This was originally written as a quick-hack with style (or lack thereof) reminiscent of /bin/mail on most Un*x machines. The whole motivation behind BM was to have some way to test the SMTP client I had written for NET and the SMTP server Phil had written and I had hacked on. I still remember how much fun it was sending the first mail message from BM to the tcp-group list... :-) And the very serious discussion Phil and I had on the phone about whether it was OK to call it 'BM' in a time period when we were trying hard to be taken seriously by the packet community at large... but hey, if you can't laugh at yourself... :-) Dave Trulli NN2Z did a bunch of work on BM at one point, which is probably why it's still alive and in use. PCELM A replacement for BM, popular in Europe. VIEW ? ucsd:/hamradio/packet/tcpip/bm One small addition: The VIEW mailer is a full screen e-mail user interface with many features like undigestifying, queue manager, gateway specification etc. NNTP readers - I have not checked yet for newreaders specific to NOS, but the ones developed for PC-UUCP use should be usable. in /ibmpc/simtel20/msdos/uucp on doc.ic.ac.uk (probably also on SIMTEL20 and its mirrors, like nic.funet.fi) It might need some modification to cooperate flawlessly with NOS NNTP but this should be minor. - snews from John McCombs - ftp.demon.co.uk:/pub/trumphurst/cppnws16.zip Nikki Locke Why does the 'shell' command just return without doing anything? This happens if there is not enough memory to run the DOS command interpreter as a child of NOS. Rather than printing a message informing you of this, NOS just prints another prompt! What is AMPRNET? AMPRNET is a network composed of amateur TCP/IP hosts whose names are in the .ampr.org domain, and whose addresses are assigned by the AMPRNET address coordinator (a person!). Although AMPRNET is technically a network, not all AMPRNET hosts can talk to one another. Many people have begun to use landline gateways to connect the disjoint pieces of AMPRNET. What is a gateway? A gateway is a host that joins one or more AMPRNET subnets to the rest of AMPRNET. Loosely, an AMPRNET subnet consists of the AMPRNET hosts in a particular area, say, a city. AMPRNET is a network composed of amateur TCP/IP hosts. In theory, any host on a network like AMPRNET should be able to send IP datagrams to any other host on the network. In practice, however, the network is fragmented into "subnets", which, loosely, is a set of machines that can talk to each other. (The term "subnet" has a more specific meaning in TCP/IP, where it refers to a set of machines that have a common prefix in their IP address. This simplifies routing; for more info, see the reference on gateways.) The hosts on a subnet may talk to each other via radio modems, TNCs, telephone modems, ethernet, wire lines, or just about any other way of moving bytes around. (NOS is one of the few packages available to amateurs that can support this diversity.) Each city or town tends to develop such a subnet, because everyone in the locality can hear one another on VHF radio, and the phone calls are local, or they're even in the same building. If each of the hosts in a locality has a coordinated AMPRNET domain name and IP address, then that subnet is technically part of AMPRNET. This means that other AMPRNET hosts on other subnets (i.e., other cities) could send packets to them, if only there were a way to get them between the two subnets. To do this, there must be at least one host that can talk to both of the subnets, through which hosts on either subnet can send their packets to the other side. Such a host is a gateway. In effect, a gateway turns the subnets into a larger network, called an "internetwork", or "internet". What is an "Internet gateway"? An "Internet gateway", in AMPRNET, is a gateway that talks to other AMPRNET gateways by sending the packets over a landline network known as the Internet. The Internet (capital I) is a world-wide academic and commercial network. Therefore, any host potentially can be a gateway if it is connected to some subnet of AMPRNET and to the Internet. In order to become an operating is, gateway, the other gateways must be told what the gateway's addressand what AMPRNET subnets it knows how to send packets to. For more information on gateways, look in the following files that are available for FTP. minnie.cs.adfa.oz.au gateways.023 (number changes with version) ke9yq.ampr.org What's the difference between a gateway and a wormhole? Wormholes don't do routing; gateways do. A wormhole is just a two-way "data pipe", often a phone line, that can accept packets on one end, and spits them out at the other end. There is only one possible destination. It is like an AX.25 digipeater, and in fact many wormholes are fashioned to look just like one. (Another way to look at it is as a band opening on HF.) A wormhole doesn't know or care what is in the packets; in particular, it doesn't do any routing. To use a wormhole, you have to know where its input port is so you can send your packets through it (as if it were a digipeater), and what stations are at the other end, so you can address packets to them. Gateways are more interesting than wormholes, because they can route packets sent into them to one of many different destinations, perhaps routing through other gateways on the way. Gateways form a true network, in the sense that you can just inject a packet into it, and it will figure out how to get it to its destination (if possible). Because they do routing, gateways have to know something about the packets you send through them, so they can figure out where to send it. That is, you have to use a particular protocol that the gateway understands. What parameters should I use for the netrom interface? The netrom interface allows you to send IP datagrams through a netrom network. It also makes your NOS station look like just another netrom node to the others. However, many people have found that using the netrom code makes NOS crash frequently, usually just after a netrom stream has been closed, especially when the connection was made by a user in the mailbox. It's best not to use the netrom code unless there is absolutely no other way you can get to other IP stations. Some people use NOS as an alternative implementation of the netrom protocol to build netrom node stacks. This usage belongs more to the realm of netrom wizardry than to NOS, so we won't discuss it here. If you're brave enough to use the netrom interface anyway, use these parameters as an example: netrom nodetimer 1800 netrom obsotimer 1800 netrom minquality 144 netrom interface axip_place 230 netrom interface diode_matrix 230 # direct rs232 netrom interface back_bone 203 # just *two* radios talking netrom interface user_port 192 # lots of radios netrom ttl 24 Since the timers will affect both the RF port and the encap ports I would suggest you set it to the same value that is used locally on the air. When somebody sends an AX.25 packet to my NOS system through a digipeater, my NOS insists on sending it back through the digipeater even though the direct path really works. How can I force it to use the direct path? Carl Makin writes: Define the route manually: ax25 route add It's then defined as a local route and should be used in preference to an "auto" route. ---------------------------------------------------------------------- Wishes Costas, SV1XV, writes: As my job forces me to run SCO Unix Sys V on my main machine, I would like to find a STREAMS/socket driver to add ax.25 and KISS to the existing SCO TCP/IP services, but it seems nobody has written one yet. ---------------------------------------------------------------------- Glossary NET NETROM NOS RSPF ucsd.edu:/hamradio/packet/tcpip/docs/rspf.doc ucsd.edu:/hamradio/packet/tcpip/incoming/rspf22p.txt RIP TCP IP Internet internet ---------------------------------------------------------------------- Bibliography ARRL Computer Networking Conference Proceedings Available from ARRL HQ, Newington CT. Send mail to info@arrl.org for an automatic response pointing at more information about the ARRL. Some of these papers are available online in the directory ucsd.edu:/hamradio/packet/tcpip/docs. This list is not exhaustive; there are many other interesting articles, but these are the ones most relevant to NOS and TCP/IP. NOS Overviews and Documentation NOS Command Set Reference Ian Wade G3NRW 10th (1991) NOSVIEW: The On-Line Documentation Package for NOS Ian Wade G3NRW 11th (1992) The KA9Q Internet (TCP/IP) Package: A Progress Report Phil Karn KA9Q 6th (1987) Amateur TCP/IP: An Update Phil Karn KA9Q 7th (1988) Amateur TCP/IP in 1989 Phil Karn KA9Q 8th (1989) Services and Protocols The Design of a Mail System for the KA9Q Internet protocol Bdale Garbee, N3EUA 6th (1987) Gerard van der Grinten, PA0GRI Finger - A User Information Lookup Service Michael T. Horne, KA7AXD 7th (1988) Callsign Server for the KA9Q Internet Protocol Package Doug Thom, N6OYU 8th (1989) Dewayne Hendricks, WA8DZP The Network News Transfer Protocol and its Use in Packet Radio Anders Klemets, SM0RGV 9th (1990) A Routing Agent for TCP/IP: RFC 1058 Implemented for the KA9Q Internet Protocol Package 7th (1988) Albert G. Broscius, N3FCT Thoughts on the Issues of Address Resolution and Routing in Amateur Packet Radio TCP/IP Networks Bdale Garbee, N3EUA 6th (1987) Another Look at Authentication Phil Karn KA9Q 6th (1987) LZW Compression of Interactive Network Traffic Anders Klemets, SM0RGV 10th (1991) PACSAT Protocol Suite -- An Overview Harold Price, NK6K 9th (1990) Jeff Ward, G0/K8KA BULLPRO -- A Simple Bulletin Distribution Protocol Tom Clark, W3IWI 9th (1990) Macintosh KA9Q Internet Protocol Package on the Apple Macintosh Dewayne Hendricks, WA8DZP 8th (1989) Doug Thom, N6OYU Status Report on the KA9Q Internet Protocol Package for the Apple Macintosh Dewayne Hendricks, WA8DZP 9th (1990) Doug Thom, N6OYU Higher Speed Amateur Packet Radio using the Apple Macintosh Computer Doug Yuill, VE3OCU 10th (1991) Network design The Implications of High-Speed RF Networking Mike Chepponis, K3MC 8th (1989) Glenn Elmore, N6GN Bdale Garbee, N3EUA Phil Karn, KA9Q Kevin Rowett, N6RCE Design of a Next-Generation Packet Network Bdale Garbee, N3EUA 8th (1989) More and Faster Bits: A Look at Packet Radio's Future Bdale Garbee, N3EUA 7th (1988) Physical Layer Considerations in Building a High Speed Amateur Radio Network Glenn Elmore, N6GN 9th (1990) Spectral Efficiency Considerations for Packet Radio Phil Karn, KA9Q 10th (1991) This should be considered to be required reading. MACA - A New Channel Acess Method for Packet Radio Phil Karn, KA9Q 9th (1990) A Duplex Packet Radio Repeater Approach to Layer One Efficiency Robert Finch, N6CXB 6th (1987) Scott Avent, N6BGW A Duplex Packet Radio Repeater Approach to Layer One Efficiency, Part Two Scott Avent, N6BGW 7th (1988) Robert Finch, N6CXB Network Implementation Packet Radio at 19.2 kB -- A Progress Report John Ackermann, AG9V 11th (1992) Implementation of a 1Mbps Packet Data Link Glenn Elmore, N6GN 8th (1989) Kevin Rowett, N6RCE Hubmaster: Cluster-Based Access to High-Speed Netowrks Glenn Elmore, N6GN 9th (1990) Kevin Rowett, N6RCE Ed Satterthwaite, N6PLO Recent Hubmaster Networking Progress in Northern California Glenn Elmore, N6GN 9th (1990) Kevin Rowett, N6RCE The 56 kb/s Modem as a Network Building Block: Some Design Considerations Barry McLarnon, VE3JF 10th (1991) Digital Networking with the WA4DSY Modem - Adjacent Channel and Co-Channel Frequency Reuse Considerations Ian McEachern, VE3PFH 10th (1991) A Full-Duplex 56kb/s CSMA/CD Packet Radio Repeater System Mike Chepponis, K3MC 10th (1991) Lars Karlsson, AA6IW A High Performance, Collision-Free Packet Radio Network Phil Karn KA9Q 6th (1987) Adaptation of the KA9Q TCP/IP Package for Standalone Packet Switch Operation Bdale Garbee, N3EUA 9th (1990) Don Lemley, N4PCR Milt Heath Hardware The KISS TNC: A Simple Host-to-TNC Communications Protocol Mike Chepponis, K3MC 6th (1987) Phil Karn, KA9Q The Ottawa Packet Interface (PI) A Syncrhonous Serial PC Interface for Medium Speed Packet Radio Dave Perry, VE3IFB 10th (1991) HAPN-2: A Digital Multi-Mode Controller fo the IBM PC John Vanden Berg, VE3DVV 11th (1992) The PackeTen system - The Next Generation Packet Switch Don Lemley, N4PCR 9th (1990) Milt Heath ----------------------------------------------------------------------