[2252] in linux-net channel archive

home help back first fref pref prev next nref lref last post

Re: NameD and Telnet

daemon@ATHENA.MIT.EDU (James G. Stallings II -- NetAdmin)
Mon Mar 25 23:20:27 1996

Date: 	Mon, 25 Mar 1996 23:01:57 +0600 (GMT+0600)
From: "James G. Stallings II -- NetAdmin" <zap@kraken.port-aransas.k12.tx.us>
To: ATPlack@scj.com
Cc: linux-admin@vger.rutgers.edu, linux-net@vger.rutgers.edu
In-Reply-To: <"MSMAIL PC */PRMD=SCJMAIL/ADMD=MCI/C=US/"@MHS>


We have a flat class C network tied to a 56K leased line via a Cisco 2501 
Router.  We run named on a 486DX100 (local bus/bus mastering fast SCSI II); 
it has 32 MB of RAM and 2GB HD.  The disk is reasonably fast (its a 
Seagate Baracuda II).  When we were running this machine as a dedicated 
nameserver (i.e. no user accounts, no extraneous processes), it ran the 
DNS so slowly that we were completely unable to connect until after 
several requests for nameservice had be made.  Lynx would only work on 
any given "new" link after attempting to activate the link several 
times.  Once the address was successfully acquired, it would remain 
cached and access to that site would become routinely quick.

I found I was able to alleviate the problem somewhat by tweaking the SOA 
time-to-live and refresh rate for the machine running the nameserver.  
This caused the system to require only a few requests for a name before 
the server was able to retrieve and cache it.  Once cached, with the TTLs 
on the local server optimised, things would crackle after the network had 
been on-line.

The single most effective thing we have done though is to add a 
cacheing-only secondary nameserver (a Sun netra i running SunOS).

If something goes down now (i.e., the primary or secondary nameservers) 
things revert to earlier states; with all of it online it fairly sizzles 
after someone pushes it out around the world for a few minutes.

I think that the language I use betrays my lack of deep technical insight 
into this matter; I feel that the limited success I have achieved has 
largely been due to some good intuition, and that I might easily screw it 
up by tweaking it further.

I do think it interesting to observe the difference that the Sun netra 
makes, and to wonder if there is something incredibly "less than optimum" 
about the port of "named" for linux.


 NetAdmin - port-aransas.k12.tx.us
<zap@kraken.port-aransas.k12.tx.us>

"Every Day, from here to there, funny things are everywhere."
  -- The Late Dr. Seuss

On 25 Mar 1996 ATPlack@scj.com wrote:

> 
> Hello,
> Interesting problem.
> 
> Once I set up named on the Linux box.  The telnet session takes 15-20 
> seconds to establish on my local Ethernet segment (using IP address and not 
> host name). (Without named, it takes less than a second.)
> 
> I have a feeling that memory could help, but before I throw hardware at it, 
> I would like to be sure.  Anyone?
> 
> 


home help back first fref pref prev next nref lref last post