[Samba] samba4 binddlz performance
Matthieu Patou
mat at samba.org
Mon Dec 10 02:13:00 MST 2012
On 12/03/2012 06:40 AM, Thomas Manninger wrote:
> -------- Original-Nachricht --------
>> Datum: Fri, 23 Nov 2012 14:32:31 -0800
>> Von: Matthieu Patou <mat at samba.org>
>> An: samba at lists.samba.org
>> Betreff: Re: [Samba] samba4 binddlz performance
>> On 11/19/2012 07:11 AM, Thomas Manninger wrote:
>>> Hello,
>>>
>>> i am using samba4rc2.
>>>
>>> I have problems with the bind9 dlz module, i get very long response
>> times from interal queries.
>>> root at s-srv01:~# dig s-srv04.test.local @192.168.0.4
>>>
>>> ; <<>> DiG 9.8.0-P4 <<>> s-srv04.test.local @192.168.0.4
>>> ;; global options: +cmd
>>> ;; Got answer:
>>> ;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 64478
>>> ;; flags: qr aa rd ra; QUERY: 1, ANSWER: 1, AUTHORITY: 2, ADDITIONAL: 1
>>>
>>> ;; QUESTION SECTION:
>>> ;s-srv04.test.local. IN A
>>>
>>> ;; ANSWER SECTION:
>>> s-srv04.test.local. 900 IN A 192.168.0.4
>>>
>>> ;; AUTHORITY SECTION:
>>> test.local. 900 IN NS s-srv01.test.local.
>>> test.local. 900 IN NS s-srv04.test.local.
>>>
>>> ;; ADDITIONAL SECTION:
>>> s-srv01.test.local. 900 IN A 192.168.0.1
>>>
>>> ;; Query time: 1239 msec
>>> ;; SERVER: 192.168.0.4#53(192.168.0.4)
>>> ;; WHEN: Mon Nov 19 16:07:59 2012
>>> ;; MSG SIZE rcvd: 108
>> .local is normally used for mdns (see.
>> http://en.wikipedia.org/wiki/MDNS#Host_Discovery), can you try with
>> another kind of tld (ie. use domain test.corp).
>>> external queries are a little bit faster:
>>>
>>> root at s-srv01:~# dig google.com @192.168.0.4
>>>
>>> ; <<>> DiG 9.8.0-P4 <<>> google.com @192.168.0.4
>>> ;; global options: +cmd
>>> ;; Got answer:
>>> ;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 56403
>>> ;; flags: qr rd ra; QUERY: 1, ANSWER: 11, AUTHORITY: 13, ADDITIONAL: 6
>>>
>>> ;; QUESTION SECTION:
>>> ;google.com. IN A
>>>
>>> ;; ANSWER SECTION:
>>> google.com. 300 IN A 173.194.35.135
>>> google.com. 300 IN A 173.194.35.136
>>> google.com. 300 IN A 173.194.35.137
>>> google.com. 300 IN A 173.194.35.142
>>> google.com. 300 IN A 173.194.35.128
>>> google.com. 300 IN A 173.194.35.129
>>> google.com. 300 IN A 173.194.35.130
>>> google.com. 300 IN A 173.194.35.131
>>> google.com. 300 IN A 173.194.35.132
>>> google.com. 300 IN A 173.194.35.133
>>> google.com. 300 IN A 173.194.35.134
>>>
>>> ;; AUTHORITY SECTION:
>>> . 45846 IN NS a.root-servers.net.
>>> . 45846 IN NS c.root-servers.net.
>>> . 45846 IN NS b.root-servers.net.
>>> . 45846 IN NS g.root-servers.net.
>>> . 45846 IN NS f.root-servers.net.
>>> . 45846 IN NS j.root-servers.net.
>>> . 45846 IN NS e.root-servers.net.
>>> . 45846 IN NS i.root-servers.net.
>>> . 45846 IN NS l.root-servers.net.
>>> . 45846 IN NS k.root-servers.net.
>>> . 45846 IN NS h.root-servers.net.
>>> . 45846 IN NS d.root-servers.net.
>>> . 45846 IN NS m.root-servers.net.
>>>
>>> ;; ADDITIONAL SECTION:
>>> a.root-servers.net. 45846 IN A 198.41.0.4
>>> b.root-servers.net. 45846 IN A 192.228.79.201
>>> c.root-servers.net. 45846 IN A 192.33.4.12
>>> d.root-servers.net. 45846 IN A 128.8.10.90
>>> e.root-servers.net. 45846 IN A 192.203.230.10
>>> f.root-servers.net. 45846 IN A 192.5.5.241
>>>
>>> ;; Query time: 281 msec
>>> ;; SERVER: 192.168.0.4#53(192.168.0.4)
>>> ;; WHEN: Mon Nov 19 16:09:06 2012
>>> ;; MSG SIZE rcvd: 511
>>>
>>>
>>> When i change to the samba4 internal dns server, i get response time
>> about ~1-2ms.
>>> But why is the bind dlz modul so slooow..?
>> you can use kcachegrind to trace bind in foreground mode in order to see
>> where the time is spent.
>>
>> Matthieu.
>>
>> --
>> Matthieu Patou
>> Samba Team
>> http://samba.org
>>
>> --
>> To unsubscribe from this list go to the following URL and read the
>> instructions: https://lists.samba.org/mailman/options/samba
> I started bind with:
>
> valgrind --tool=callgring /usr/sbin/named -c /etc/bind/named.conf -f
>
> So i get any answers, bind is very slow.
Yeah callgrind tends to slow down a lot but it's super useful.
>
> Now, i have a callgrind file, but i dont can read this file... I only see, that ltdb_search_indexed needs incl. 96%..
>
> can somebody helps me??
>
> file is included as attachment.
>
> thanks!
Sorry I'm just able to have a look at it now, so yes most of the time is
spent in the ltdb_search_indexed.
Looking at the code and the capture in the dlz_bind9.c it seems that the
function dlz_lookup_types is call 44 times and is doing 44 ldb_search
with a base scope on a given DN.
Those 44 search yield 144 calls to the ltdb_search and given the fact
that we ends up with 147862 calls to ltdb_search_dn1 I'm highly sure
that not all the internal searches where done with a base scope
(otherwise we would have 144 calls to ltdb_search_dn1).
The initial search for me looks perfect and there is no real possibility
to make it any better scope=base + specified dn is the ideal case imho.
I'm looking at a small optimization in the LDB indexing to save DNs in
the index already case_folded which should help quite a lot as you spent
half of the time doing this kind of operation.
Another way to improve this is to understand why we have a spread of
ltdb_search calls and why some of those calls seems to be with a subtree
scope.
Could you rebuild the library with make clean; ./configure.developper;
make -j and is it possible to share your database ?
Thanks.
Matthieu
--
Matthieu Patou
Samba Team
http://samba.org
More information about the samba-technical
mailing list