HAM Radio 1/HamRadio.cdr/tcp/grinost/readme03.txt

Text File | 1991-06-26 | 5.4 KB | 130 lines

June 26, 1991

By Mike Bilow, N1BEE,
For GRINOS/GRINOS-S KA9Q 910605/PA0GRI 910619v1.6h/N1BEE 910626v0.3ß


     All users of GRINOS/GRINOS-S are invited to join the grinos@n1bee
mailing list.  This is what I use to distribute advisories and updates.
There is no requirement that you actually use GRINOS in order to join the
mailing list.  Just send a message to n1bee@n1bee and you will be added.


     I. KA9Q/PA0GRI changes

     This release of GRINOS uses KA9Q 910605 as its base instead of 910531.
The new base NOS fixes a couple of problems in the ENET driver, which is
not included in GRINOS-S.  There is a bug fix in the IP routing code, and
RSPF is believed to have been fixed; both of these are included in both
GRINOS and GRINOS-S.

     MS-DOS 5.0 is now fully supported.  NOS uses an undocumented internal
MS-DOS structure, and the "status" command and a few other things did not
work right in the old base NOS.  Since PA0GRI did the fix in his release by
incorporating the N1BEE code, users of GRINOS since my 910612v0.2ß release
will see no difference.


     II.  N1BEE changes

          A.  Bug fixes and default modifications

     The "info" command now displays the actual number of sockets created
rather than the default.  (Reported by KA1XN.)

     The "memory threshold" parameter defaults to 8192.  It is not recommended
that this be increased, as it will force garbage collection.  The garbage
collection code is not right, and will result in memory corruption, which
will either lock up the machine (with "watchdog off") or reboot it (with
"watchdog on").  Setting "watchdog on" is recommended.  (Reported by KA1XN
and KA1MF.)

     DRSI users can get proper tracing of output frames by increasing their
"memory nibufs" to about 9 to 12, depending upon their machine speed.  The
objective should be to set "memory nibufs" sufficiently high so that the
"memory status" display shows "Ibuffail" at 0 and "Iminfree" just above 0,
ideally around 2.  DRSI users can safely set "memory ibufsize" to 1024,
which is half of the 2048 default.

     Non-DRSI users will likely see no counts for "Interrupts-off calls"
to alloc and free on their "memory status" display, in which case "memory
nibufs" and "memory ibufsize" do not matter.  Settings as low as "memory
ibufsize 512" and "memory nibufs 3" are probably safe, and will save about
10K of memory over the defaults.  (DRSI is not available in GRINOS-S.)

     The "tcp irtt" now defaults to 15 seconds instead of 5, which is far
more realistic.  Most users override the default anyway.

     The "tcp timertpye" now defaults to linear instead of exponential.


          B.  New features

              1.  Hard limit for tcp backoff

     There is now a "tcp blimit" command.  It works in a similar way to the
one provided by KA1JY in his modifications to KH113016 NOS, but it is a bit
different and was recoded from scratch.  (I don't have his source code.)
The "tcp blimit" is a small constant between 1 and 31, and represents the
maximum backoff count that will be allowed.  In general, the maximum time
of a backoff for a given tcp channel will be (2 * tcp blimit * tcp rtt) if
linear backoff is used, or (tcp rtt * 2 ^ tcp blimit) if exponential.

     The rtt (round trip time) is measured for each tcp channel and is
adjusted dynamically during the connection.  As a result, although "tcp
blimit" creates a hard limit, it will be different for each channel.  Users
are warned not to set "tcp blimit" to 0, although the code will accept that,
as it will completely stop the timers from running so that lost frames will
never be retried.  (This is allowed for experimentation.)

     This command will allow fairly arrogant and obnoxious domination of a
channel, and should be used with extreme care, especially if a lot of the
frames being retried over and over are SYN frames for hosts not on the air.
If using linear backoff, values of "tcp blimit" below 16 are not recommended;
this would correspond to a hard limit of about 5 minutes.


               2.  Domain translation

     As explained in the new file DOMAIN.DOC, GRINOS observes different rules
for translation of host names to IP addresses than older NOS versions such as
KH113016.  The "domain trace on" command has therefore been enabled for those
trying to analyze problems with DOMAIN.TXT files.  It existed before, but it
did not do anything.


     III.  Remaining known problems

     The ttylink system is seriously messed up.  It is functional, but the
screen I/O gets extremely confused.  I am working on this.

     The memory allocator still does not work, despite repairs to a couple
of identifiable problems.  Any garbage collection will likely result in
"invalid free" error messages.  Any counts of invalid frees and the like
seen with the "memory status" display should be reported.  As noted, keep
"memory threshold" low to minimize this, and do not shell out to DOS or
the mailer until the heap has stabilized, indicated by the "more cores"
counter not incrementing.


     IV.  Current mailing list

     Here is the current definition for grinos@n1bee:

    n1bee
    [e-mail]
    [e-mail]
    [e-mail]
    [e-mail]
    [e-mail]
    [e-mail]
    [e-mail]
    [e-mail]
    [e-mail]
    [e-mail]
    [e-mail]
    [e-mail]
    [e-mail]
    [e-mail]