Logfile on changes in the Solaris 2.x tuning document set
I'm afraid you have to read German date formats: DD.MM.YY. On popular demand
I created this changelog file for your convenience. With the help of this
changelog file, you can determine recent changes at one glance. If a file
changed without an entry here, only spelling and other minor mistakes were
Sean O'Neill made me (consciously) aware of the
Changed several typographical errors on the
index page, kindly pointed out by
Joel dot Shandelman at optionable dot com Joel Shandelman.
Added recommendation not to lower
tcp_ip_abort_interval, as this might result in Veritas backup failures amongst other things: socket write error. Pointed out by tanvir at mail dot orgTanvir Hassan, thanks.
Added maxphys, pidmax,
reserved_nprocs, physmem, and
Changed your-tune scripts to nettune
script to maintain compatibility with YaSSP.
Moving to my own domain sean.de.
Yermilov Sergey provided me with the URI of the SUN Tunable Parameters
Added a section to the ethernet problems which describes how to
enable the distinct ethernet MAC address for each interface.
dave at srce dot hrDrazen Kacar pointed
out the new Solaris 8 6/00 features concerning the DNLC:
dnlc_dir_max_size, dnlc_dir_min_size and
Solaris 8 has a default soft limit of 256, pointed out by
dave at srce dot hrDrazen Kacar.
Also fixed some of the wording for the soft- and hard limits.
The 64 bit ABI stacks do not include the execute bit by default.
The fixes to noexec_user_stack were pointed
out bydave at srce dot hrDrazen Kacar.
Added nfssrv:nfs_portmon to the descriptions of
/etc/system variables, kindly pointed out by Mr. Amos
(HK Hosting Ltd.).
Added more emphasize to the fact that TCP watermarks will be rounded
up to the nearest MTU. jason at pattosoft dot com dot au
Jason Patterson pointed out that using a maximum
of 65535 will be rounded up to 66608, thus silently turning on the
window scaling which is incompatible with IP-Filter.
Removed some outdated Squid-1 remarks. Nobody should use a Squid-1
Corrected some naive assumptions in the archaic section on
tcp_conn_req_max, brought up by Demian Hanks.
Updated the information for pt_cnt according
to the Solaris 8 Admin Answerbook. Pointed out by dave at srce dot hr Drazen Kacar.
Added a pointer to the Solaris FAQ to the literature section.
Blair Zajac (author of ORCA, a webserver
monitor) provided corrections for over fifty misspellings, or
over hundred, if you count AE and BE differences as mistakes. Which
makes me think, should I fashion my style more to AE or more to BE?
Michael Fromme forwarded an
email from the multimedia conference list, stating that early ethernet
interfaces on low versions of Solaris were only capable of participating
in 64 multicast groups simultaneously.
Inserted a new chapter
What's new containing a few of the most exciting new features in
Added a note to tcp_deferred_ack_interval
which mentions its influence on slow bulk data transfer. Shobana
Narayanasw (MIL3) had me look into the matter.
Jens Elkner again:
Fixed rlim_fd_* notions
about breaking the license server and RPC. Both services only look at
the softlimit; the hardlimit may be increased without harm.
Shobana Narayanasw (MIL3) made me aware of the fact that my
documentation was mistaken. Actually, the ACKs max parameter does
influence bulk data transfer, and thus limits the number of segments
for which an ACK may be outstanding.
What started as fixes on the design actually enhanced the meta
tags, fixed dangling and missing end-tags. The general appearance
was reworked to be more consistent.
The framed version was scratched due to the frame-inherent difficulties
with the title bar and the overhead for superflous design gimmicks.
One CSS1 conforming stylesheet was introduced, but only for visual
enhancements, as even Netscape 4.x does not print what is declared in
the stylesheet. And a printed document with all structuring tags removed
is virtually worthless.
Have a guess... I love motorbikes ;-) Click it in a spare
Many minor changes, partially in sequence of paragraphs, partially
Reworked externally pointing hyperlinks to be shown in one
frame of their own.
reworked the introductive section to
contain more information on the use of
ndd. Also moved the table of
contents to the very beginning, and the introductory words into the
Changed the ndd.pl script for examining values
to contain a help, and fixed the tabular values.
Refreshed links to the HTTP/1.1 protocol from 3rd to 4th revision.
and noexec_user_stack_log to the
/etc/system section, in accordance with CA-98.06. Be warned
to try out this option thoroughly before using it on a production
Exchanged section 7.3 and 7.4, since readers told me about "using ndd
to display hme values", even though I mentioned it someplace later. Now I
want to point directly (in the next section) to ndd
for extracting current
information from network driver loadable kernel modules.
Added a shell script to displays some
/etc/system tunable kernel parameters to the
scripts section. The script was kindly
supplied by andre at online dot ee Andres Kroonmaa.
Added the 10 Mbps downgrade hme_adv_10fdx_cap and
hme_adv_10hdx_cap options to the
fast ethernet section, as several readers
mentioned them, Mr. Kannemann among them.
Fixed a few quite severe bugs with the first
startup script, pointed out and fixes suggested by
chouanard at parc dot xerox dot com Jean Chouanard
ip_respond_to_echo_broadcast to the
common parameters section, also pointed
out by Jean Chouanard.
Worked on tcp_slow_start_initial relevant sections,
initiated by mbennett at netcom dot com Mike Bennett.
Found and added the explaination for *_max_buf.
Just added Sun's public TCP/IP and Data Communications Administration Guide to the literature section, kindly pointed out by Richard Murphy (SwRI). Especially the high-speed TCP stuff needs some more work in my document.
Added reasons for the recommended buffer sizes.
Added TODO item to look into possible NAGLEing of HTTP requests.
Fixed mispelled maxusers, caught by
jw at wede dot de Jan Wedekind (UUNET Germany).
Updated the startup script not to assume that
/var is mounted at the time the script is called. Now will work
on more Solari than before, and does verbose output. Solaris < 2.5.1
will need to modify the patch-finder part of the script.
Corrected my statements to the Solaris 2.6 parameter
to experimentations with this parameter. The information is still not
Added new section of of things in need
to be done (TODO list). The section is a meta-section covering
intend to dig into. Thus you get ample warning about things which might
change in the foreseeable future.
Added the use_mxcc_prefetch parameter.
Removed advice to fiddle with maxusers.
sfx at unix-ag dot org Lars Eilebrecht made me aware of
a the formula concerning ufs_ninode and
He also pointed out the default behaviour of bufhwm and
Adrian Cockroft reported that his new book which is due to appear
on 10th of April 98 clarifies some of my open issues. Added to the
literature section. I expect it will correct many errors I made along
the way, too.
Changed example startup script: Moved sections
around a little and added tcp_slow_start_initial.
fixed many bugs along the way, especially on the SYSV IPC page.
updated the patchlist.
updated hints to the TCP patchlist.
Documented new parameters tcp_conn_req_max_q and
tcp_conn_req_max_q0. They seem to spring into
existence after applying patch 103582 > pl13. The new parameters
replace the extensive description of tcp_conn_req_max.
Corrected spelling of tcp_ip_abort_cinterval.
Corrected semantical errors in the documentation of
tcp_ip_abort_interval, referenced SYN flood attacks.
Got the first indication of a kernel patch (yes, this is still
possible with Solaris) increasing the hard limit of
tcp_conn_req_max above 1024.
Documented parameter hme:hme_adv_autoneg_cap
as seen with Mr. Hüsemann
Improvements on the SYSV IPC parameters. Added a small program
to dynaload the necessary streams modules into the kernel.
Clarifications on the maximum number of possible file descriptors.
Test programs for processes and file descriptors. Yes, there are
other ways to find out about those, but why not just do the very same
thing your server do when they die unexpectedly.