How frequently Bind DNS can be updated?

[Mark_Andrews at isc.org]

> Hi,
> 
> First of all Thanks a lot for your response, Simon.
> 
> I have few more questions which came after going through your reply.
> 
> You mentioned about the log file which DNS maintains. I belive the =
> number of operations that can take place per second is not only hold =
> good for Update, but this holds good for all operations.
> 
> Correct me if Iam wrong, I belive we should be able to do some X number =
> of transactions with DNS per second, and in your case this X is 10. The =
> operation can include Query/Delete/Add.

	Updates require disk i/o.  Each succesful update requires the i/o
	to have completed before success is returned.

	Queries are memory only operations.  If named is forced to page
	the queries/second drops dramatically.  Normally it is in the
	thousands of queries/sec range.
 
> Because I concluded that the problem Iam facing could be due to this =
> restriction, as in my case there will be more than 10 operation =
> happening, which includes Query,Addition and Deletion.

	The DNS really is not the right place to be doing sub-second load
	balancing.
 
> Is there any site or resource available which gives the details about =
> such a performance figures about Bind DNS ?? If it is there, please pass =
> that to me.
> 
> - vinod=20
> 
> 
> 
> -----Original Message-----
> From: Simon Waters [mailto:Simon at wretched.demon.co.uk]
> Sent: Sunday, September 07, 2003 4:35 PM
> To: comp-protocols-dns-bind at isc.org
> Subject: Re: How frequently Bind DNS can be updated?
> 
> 
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Vinod Kumar wrote:
> >
> > Iam suspecting whether "there is any restriction" from bind that DNS =
> =3D
> > shouldn't be updated very frequently ??
> 
> As far as I know BIND 9 is hardware limited in this respect.
> 
> I assume a bog standard PC, or similar, with one disk in is going to be
> stressed if you maintain sending much more than 10's of updates a
> second, purely because it logs, and by default adds data to the system
> log, for each update, on one disk this will pretty quickly cause issues.
> 
> Have you got the log from when the problem occurred, as this should
> record what updates happened.
> 
> I haven't seen anyone stressing out BIND via dynamic updates before, but
> most people using dynamic update are using it to mimic NetBios names
> over NetBeui using IP networking, and it wouldn't be hard to exceed the
> reliability of the system being replaced. Most of the new Microsoft
> stuff achieves reliability by perpetually reannouncing itself, which I
> suppose is one way of doing things.
> 
> You haven't mentioned what is happening at slave servers.
> 
> You can limit the transaction log file size in BIND 9 which would appear
> to impose a different possible mode of failing.
> 
> BIND 8 appears to have had extra limits related to IXFR, but I've never
> seen the point in using BIND 8 for dynamic DNS work.
> 
> I would have thought the main point of SRV records weight field is you
> don't need to send lots of DNS updates to balance load, so if it causes
> a problem, don't do it as often.
> -----BEGIN PGP SIGNATURE-----
> Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
> 
> iD8DBQE/WxC9GFXfHI9FVgYRAnoBAJ0YzK3iJcHUVGwLzC4A11AHzcetpQCeOExU
> tHPXYFwrem+0Mtik686DyLM=3D
> =3DLejP
> -----END PGP SIGNATURE-----
> 
> 
> 
--
Mark Andrews, Internet Software Consortium
1 Seymour St., Dundas Valley, NSW 2117, Australia
PHONE: +61 2 9871 4742                 INTERNET: Mark.Andrews at isc.org