Yes, I know: Set-based ACLs are slow even if one properly sets index configuration.
But I'd like to raise a discussion here whether it would be possible to improve the performance within slapd.
Any thoughts?
Ciao, Michael.
On 02/03/2014 07:48 PM, Michael Ströder wrote:
Yes, I know: Set-based ACLs are slow even if one properly sets index configuration.
But I'd like to raise a discussion here whether it would be possible to improve the performance within slapd.
Any thoughts?
What about profiling some use cases of interest?
p.
Our public ldap server - ldap.surevoip.co.uk makes heavy use of set so I'd be interested in sponsoring any work.
Thanks.
On 3 February 2014 19:22, Pierangelo Masarati pierangelo.masarati@polimi.it wrote:
On 02/03/2014 07:48 PM, Michael Ströder wrote:
Yes, I know: Set-based ACLs are slow even if one properly sets index configuration.
But I'd like to raise a discussion here whether it would be possible to improve the performance within slapd.
Any thoughts?
What about profiling some use cases of interest?
p.
-- Pierangelo Masarati Associate Professor Dipartimento di Scienze e Tecnologie Aerospaziali Politecnico di Milano
Pierangelo Masarati wrote:
On 02/03/2014 07:48 PM, Michael Ströder wrote:
Yes, I know: Set-based ACLs are slow even if one properly sets index configuration.
But I'd like to raise a discussion here whether it would be possible to improve the performance within slapd.
Any thoughts?
What about profiling some use cases of interest?
I'm not sure what exactly you mean. Be assured that I'm already trying to re-arrange the ACLs to make things faster. IMO I also defined decent indexing configuration.
Or do you want to see some use-cases I've implemented?
Ciao, Michael.
On 02/03/2014 10:12 PM, Michael Ströder wrote:
Pierangelo Masarati wrote:
On 02/03/2014 07:48 PM, Michael Ströder wrote:
Yes, I know: Set-based ACLs are slow even if one properly sets index configuration.
But I'd like to raise a discussion here whether it would be possible to improve the performance within slapd.
Any thoughts?
What about profiling some use cases of interest?
I'm not sure what exactly you mean. Be assured that I'm already trying to re-arrange the ACLs to make things faster. IMO I also defined decent indexing configuration.
Or do you want to see some use-cases I've implemented?
I mean apply realistic load to realistic set-based ACLs and profile the execution, to track possible bottlenecks, to avoid spending time in optimizing code that doesn't get used.
p.
--On Monday, February 03, 2014 10:30 PM +0100 Pierangelo Masarati pierangelo.masarati@polimi.it wrote:
Or do you want to see some use-cases I've implemented?
I mean apply realistic load to realistic set-based ACLs and profile the execution, to track possible bottlenecks, to avoid spending time in optimizing code that doesn't get used.
i.e, use something like oprofile for profiling the actual code execution of your set ACLs.
--Quanah
--
Quanah Gibson-Mount Architect - Server Zimbra, Inc. -------------------- Zimbra :: the leader in open source messaging and collaboration
Quanah Gibson-Mount wrote:
--On Monday, February 03, 2014 10:30 PM +0100 Pierangelo Masarati pierangelo.masarati@polimi.it wrote:
Or do you want to see some use-cases I've implemented?
I mean apply realistic load to realistic set-based ACLs and profile the execution, to track possible bottlenecks, to avoid spending time in optimizing code that doesn't get used.
i.e, use something like oprofile for profiling the actual code execution of your set ACLs.
One thing that definitely strikes me is that there are so many "not indexed" messages logged.
In my setup there are ~40..60 stupid index warning messages sent to syslog for a single search!
=> ITS#7796
Ciao, Michael.
-
Gavin Henry -
Michael Ströder -
Pierangelo Masarati -
Quanah Gibson-Mount