Re: poor performance of OpenLDAP vs AD?

[Igor Brezac <igor@ipass.net>]

On Thu, 14 Jul 2005, Christopher Hicks wrote:

> On Thu, 14 Jul 2005, Igor Brezac wrote:
> > On Wed, 13 Jul 2005, Quanah Gibson-Mount wrote:
> > > Qualitative statements about the general performance of any set of given 
> > > Directory Servers can be made if you make everything "the same" as much as 
> > > is feasibly possible.
> >
> > I do not see this being realistic.  You have fallen short in this regard 
> > with your Symas benchmarks.  Yes, benchmark results can be used as a 
> > general performance indicator, but one should not assume that their ldap 
> > deployment will perform the same.  We all have our own requirements, 
> > different machines, schema, data set size, indexes, etc., and more than 
> > likely they are different than what is used in benchmarks.
>
> So?  If the benchmark is a responsibly executed test it is reasonable to 
> presume that faster hardware will perform at least slightly better and that 
> worse hardware will perform at least slightly worse.  Doing these things 
> responsibly doesn't lead to them being done perfrectly so we are all free to 
> repeat the test ourselves if they seem wrong or don't fit our situation well 
> enough.

This is my initial point.  We should perform our own benchmarks.

> But extrapolating from past experience to future expectations is 
> part of what IT profressionals do regulalrly.  Benchmarks are but one point 
> of data to use in such extrapolations, but thank goodness good benchmarks 
> exist to give us some open yard stick to compare our own results to.

I agree.

> However, benchmarks are most often used to choose one thing over another.  If

I am glad you said that.  One more reason to perform your own benchmarks 
with your own data.

> speed is a critical consideration, changing the other variables may not 
> matter much in the scheme of things.
>
> If you have a beef with the Symas benchmarks consider publishing some of your 
> own with an explanation of why they're better.  Maybe you have some 
> substantive beef with their benchmarks, but it is not apparent to me from 
> what you've said here.

> From Quanah: "However, if you set up your benchmarks where you document 
well what you did, and make sure that your configuration
settings are public, you can set up benchmarks that others can 
duplicate, and can be agreed upon as indicative of real meaning."

Do not get me wrong, Symas provides valuable information, but it can do 
more: provide DirectoryMark configuration, data set file, slapd.conf, 
commands used to perform tests, etc.  I would certainly share my findings 
with Symas and perhaps they can publish user contributed benchmarks.

--
Igor