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 (athena.apache.org: domain of elecharny@gmail.com
 designates 209.85.215.177 as permitted sender)
Message-ID: <51926A4D.9080900@gmail.com>
Date: Tue, 14 May 2013 18:46:05 +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/20130328 Thunderbird/17.0.5
MIME-Version: 1.0
To: Apache Directory Developers List <dev@directory.apache.org>
Subject: Re: comparison of performance between JDBM and Mavibot
References: 
 <CABzFU-cDqCd9D-12ZpdxT0Vsco0qSViVkk8eJOXbtEhebo2e3A@mail.gmail.com>
In-Reply-To: 
 <CABzFU-cDqCd9D-12ZpdxT0Vsco0qSViVkk8eJOXbtEhebo2e3A@mail.gmail.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Le 5/14/13 3:22 PM, Kiran Ayyagari a écrit :
> Hi guys,

Hi Kiran,
>
>      Below given are some results of performance tests ran with jdbm and
> Mavibot
>
>     The specs of the test system are:
>
>                OS: Ubuntu Linux, 32 bit
>              RAM: 4GB
>              CPU: Intel core2 duo 1.4GHz
>              HDD: SAMSUNG SSD Thin uSATA 128GB M
>         FS type: ext4
>              Java: 1.6.0_26 (HotSpot)
>
> all tests were ran with default JVM parameters
>
> SearchPerfIT -
>           SUB level - Delta : 26530( 37690 per s ) /42680, count : 1500000
>           OBJECT level - Delta : 25253( 39599 per s ) /38524, count :
> 1500000
>           ONE level - Delta : 14642( 34145 per s ) /22352, count : 750000
>
> SearchPerfIT(mavibot) -
>           SUB level - Delta : 13079( 76450 per s ) /22880, count : 1500000
>           OBJECT level - Delta : 21506( 46498 per s ) /32397, count :
> 1500000
>           ONE level - Delta : 7916( 63160 per s ) /12214, count : 750000
>
> GetRootDsePerfIT - Delta getRootDSE: 6115( 163532 per s ) /9513
> GetRootDsePerfIT(mavibot) - Delta getRootDSE: 6087( 164284 per s ) /9594
>
> hasEntryPerfIT - Delta hasEntry: 1574( 63532 per s ) /3254
> hasEntryPerfIT(mavibot) - Delta hasEntry: 1274( 78492 per s ) /2500
>
> LookupPerfIT -
>           Delta rootDSE all attrs : 9687( 103231 per s ) /15412
>           Delta all user attrs: 13254( 75448 per s ) /20512
>           Delta all attrs: 13460( 74294 per s ) /20289
>           Delta Authz : 12157( 82257 per s ) /18712
>           Delta rootDSE all user attrs : 9855( 101471 per s ) /14835
>
> LookupPerfIT(mavibot) -
>           Delta rootDSE all attrs : 9500( 105263 per s ) /15177
>           Delta all user attrs: 9708( 103007 per s ) /14839
>           Delta all attrs: 9639( 103745 per s ) /14550
>           Delta Authz : 8863( 112828 per s ) /13517
>           Delta rootDSE all user attrs : 9695( 103145 per s ) /14575
>
> AddPerfIT - Delta : 1802698( 5 per s ) /2667678
> AddPerfIT (mavibot) - Delta : 54907( 182 per s ) /90782
>
> ComparePerfIT - Delta compare: 1921( 52056 per s ) /5062
> ComparePerfIT(mavibot) - Delta compare: 1868( 53533 per s ) /4275
>
> DeletePerfIT - Delta deletion: 2294007( 4 per s ) /3485085
> DeletePerfIT(mavibot) - Delta deletion: 72569( 137 per s ) /113611
>
>
This is impressive !

I'm a bit surprie by the extremely low performance on write. On my
computer (MacBook pro, 8Gb, SSD 256Gb, Core2Duo 2.8 Ghz) I can add up to
80 entries per second. If I do the math correctly, I woud get more than
2500 additions per second on my cmputer :-)

Ok, we know that Mavibot is really young, and that we still have to
reclaim the free pages, which will add some strain on the CPU (but not
that much). Currently, the file will grow insanely, as we keep all the
versions until we restart the server, something we simply can't let run
in production.

Anyway, this is just a great result, very promising for the near future !

Also the good point is that Mavibot allows concurrent reads with write -
but not concurrent writes - when the current implementatin (using JDBM)
serializes every read when a wite is running : a real contention.

Many thanks for the hard work, Kiran !

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