[svlug] Reprise: Adjusting retransmission parameters

Karen Shaeffer shaeffer at best.com
Wed Feb 21 00:49:01 PST 2001


On Tue, Feb 20, 2001 at 06:26:46AM -0800, Eric Nickell wrote:
> 
> I've looked at each of the variables and had the guy here at my office
> who is the Sun networking guru scan it as well.  I am, unfortunately,
> ignorant of the most basic understanding of tcp/ip.  In any case,
> looking through the documentation variable by variable, I did not see
> anything that looked like the obvious thing to try.  A howto on the net
> also suggested that it was a bit dangerous to go poking values
> indiscriminately into the networking stuff. (Dangerous in the sense of
> leaving your machine in an unbootable state.)


Nonsense. Just get a rescue floppy and remember what you change. Besides,
those /proc parameters are reset when you reboot. Now, if you go into the
header files and make modifications (The next step in the game.) Then you
definitely want to have a rescue disk handy.

Go ahead and experiment. (That's what I would do. This _is_ Linux, after
all.) Oh, and please let us know what you learn.

> I'm afraid that I cannot produce questions with more specificity.
> Possibly, is there further documentation on any of these variables, esp
> those marked "undocumented" in the group that includes tcp_retries1,
> tcp_retries2, tcp_max_delay_acks, tcp_fin_timeout, tcp_max_ka_probes,
> tcp_hoe_retransmits, which are listed as "Undocumented for now"?
> 

TCP/IP documentation? Well, the Richard Stevens books are very well done.

TCP/IP Illustrated Volume 1, The Protocols
TCP/IP Illustrated Volume 2, The Implementation

And then there are all the RFCs as well.


> It's looking bleaker and bleaker...

Well, if you experiment, you will definitely have some fun, learn some
stuff, and maybe even tweak out some performance gains. As much as I would
like to look into this--I'm quite busy at the moment.

No doubt you have already completed some testing using traceroute, ping,
tcpdump, netstat, and even strace. You can learn quite a lot about the
dynamics of your problem using these tools.

c,
-- 
 Karen Shaeffer
 Neuralscape; Santa Cruz, Ca. 95060
 shaeffer at neuralscape.com  http://www.neuralscape.com




More information about the svlug mailing list