Mailing-List: contact dev-help@directory.apache.org; run by ezmlm
Precedence: bulk
Reply-To: "Apache Directory Developers List" <dev@directory.apache.org>
Received-SPF: pass (nike.apache.org: domain of elecharny@gmail.com designates
 209.85.212.180 as permitted sender)
Message-ID: <52169A50.6020007@gmail.com>
Date: Fri, 23 Aug 2013 01:10:08 +0200
From: =?UTF-8?B?RW1tYW51ZWwgTMOpY2hhcm55?= <elecharny@gmail.com>
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.7;
 rv:17.0) Gecko/20130801 Thunderbird/17.0.8
MIME-Version: 1.0
To: Apache Directory Developers List <dev@directory.apache.org>
Subject: Re: ApacheDS performances
References: <5216852E.10004@gmail.com>
 <F6DD3E4E6048AA4B5308220B@[192.168.1.22]>
In-Reply-To: <F6DD3E4E6048AA4B5308220B@[192.168.1.22]>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Le 8/23/13 12:01 AM, Quanah Gibson-Mount a écrit :
> --On Thursday, August 22, 2013 11:39 PM +0200 Emmanuel Lécharny
> <elecharny@gmail.com> wrote:
>
>> 1) OpenLDAP is the fastest, with at least 26 000 search/s, and the CPU
>> wasn't maxed (we reached 85%). The network transfer rate was around
>> 20Mb/s.
>
> As a reference, it takes me about 120 clients running on 8 different
> servers (15 clients per server) driving load against OpenLDAP to max
> out slapd at 53,000 searches/second.  8 clients gets me 11,597
> searches/second, 16 clients is 21,754 searches/second etc.  It takes a
> lot of clients to really understand the full performance profile of
> OpenLDAP.

We plainly understand that ! Especially with the client we are using,
which is doing a lot of un-needed CPU consumming tasks (like decoding
the response, etc). At some point, striping out the client from what is
spurious for a test would help us a lot.

We started to run the test with one machine, and went up to 15 000
search/s (33% CPU on the OpenLDAP server, 100% CPU on the client). We
had 20 threads running on the client.

We thrown 3 more injectors, and 30 more threads, to reach barely 63% CPU
on the server (and 23 000 request/s). With a fifth client, 20 threads,
we maxed at 85%. At this point, we ran short of injectors (5, for a
total of 80 threads)!

We hope to be able to run the benchmark on another platform at the
beginning of september.

Thanks !

-- 
Regards,
Cordialement,
Emmanuel Lécharny
www.iktek.com