Re: DNS servers
On Fri, 22 Nov 2002 22:58, Toni Mueller wrote:
> > LDAP or SQL backed DNS isn't an option unless performance is not
> > required. A LDAP or SQL query takes far longer than I want my DNS
> > lookups to take.
> Here I'd like to re-use the words of DJB: "Profile, don't speculate."
> Apart from the fact that LDAP (and SQL) performance varies wildly
> across different servers - eg. Fefe once claimed that his LDAP server
> ran several orders of magnitude faster than OpenLDAP at a time, and
> in a special situation that was important for him - we already know
> about tinydns' ability to serve some 6000 requests per second on
> decent dual cpu PC hardware, and we also know that on average, the
> ldapdns by Mrs. Brisby runs twice as fast as tinydns using OpenLDAP.
> This software serves it's data directly from the LDAP backend to the
> best of my knowledge - having no intermediate format was a design
> goal. How fast do you need to get?
Last time I benchmarked OpenLDAP for user authentication (basically checking
user names and comparing the hashed password strings - nothing advanced) it
didn't deliver nearly that performance.
I've spent quite a bit of time benchmarking OpenLDAP in various ways, last
time I was testing it I found the performance much lower than I wanted from a
DNS server. I guess that if the DNS server agressively caches the data it
gets from LDAP then performance should be fine though.
> > Of course that plan doesn't work so well if you are hired by a company
> > that doesn't see the value of a lab and provides no decent resources for
> > testing.
> Hmmm... A company that has no idea of the value of a lab???
> > There was one time I was setting up some fully loaded E4500 machines as
> > LDAP servers and I had to use my Thinkpad for some tests because there
> > was nothing else that I could use. A Thinkpad running Linux is not much
> > good for testing the client and server sides of an operation that will be
> > deployed on an E4500, but it was the best I had.
> Ok, define 'lab'...
> Having some spare equipment that can be used to set up experimental
> networks to check things out is not only a basic business requirement,
> but also (mostly) cheap.
True. But you need space for it, electricity, and it produces noise (humming
of cooling fans etc). Most companies would rather give the space to managers
or marketting people. Allocating 3 times the space per person to the
technical team that other teams get so that they can run a lab is something
that usually doesn't happen.
http://www.coker.com.au/selinux/ My NSA Security Enhanced Linux packages
http://www.coker.com.au/bonnie++/ Bonnie++ hard drive benchmark
http://www.coker.com.au/postal/ Postal SMTP/POP benchmark
http://www.coker.com.au/~russell/ My home page