Mailing-List: contact user-help@couchdb.apache.org; run by ezmlm
Precedence: bulk
Reply-To: user@couchdb.apache.org
Received-SPF: pass (athena.apache.org: domain of adam.kocoloski@gmail.com
 designates 209.85.160.180 as permitted sender)
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0 (Apple Message framework v1084)
Subject: Re: CouchDB performance
From: Adam Kocoloski <kocolosk@apache.org>
In-Reply-To: 
 <CAJzZBASjfdUVgmY14zCq9MC0+cf24yUzvfGhRAxoMOtw7TchMg@mail.gmail.com>
Date: Thu, 11 Aug 2011 14:02:03 -0400
Content-Transfer-Encoding: quoted-printable
Message-Id: <9E9A84DE-CA6F-453B-9FDC-C5BCA48F0B24@apache.org>
References: 
 <CAJzZBASjfdUVgmY14zCq9MC0+cf24yUzvfGhRAxoMOtw7TchMg@mail.gmail.com>
To: user@couchdb.apache.org

Hi John, first thing I'd point out is that Filipe and Paul have done =
quite a lot of work on the JSON processor since 1.0.2 was released.  =
Trunk contains a JSON processor that calls out to YAJL instead of =
relying on pure Erlang code.  The improvement for large document bodies =
is pretty substantial.  Also, processing MBs of JSON is going to be =
CPU-bound in 1.0.2.  CouchDB makes good use of multicore servers to =
distribute some of the load if you throw concurrent requests at it.  =
Hope that helps,

Adam

On Aug 11, 2011, at 1:54 PM, John Cheng wrote:

> I published some Python scripts to help me evaluate the performance of
> CouchDB (https://github.com/jlcheng/couchdb-test). I was hoping the =
folks
> here can help me understand what I am seeing.
>=20
> The raw spreadsheet of my findings:
> =
https://spreadsheets0.google.com/spreadsheet/pub?hl=3Den_US&hl=3Den_US&key=
=3D0AtagEcs-wlFDdGVfMGItdUIxQXcyTmZxdVdOMUw4TWc&output=3Dhtml
>=20
> First, I've found that the Document Update Handler API performed worse =
than
> using Python to get, modify, then update the document when the =
application
> code is running on the same machine as CouchDB. This is surprising to =
me. It
> leads me to believe that when the network is not a bottleneck, it =
might be
> better to avoid update handers.
>=20
> Second, the insert performance is not as good as I'd hoped - at least =
for
> large documents (100 kB+). The update performance is even worse. When =
using
> the "benchbulk.sh" script with a DOCSIZE of 100 kB and a BULKSIZE of =
15, I
> am seeing about 12 inserts per second. Using my Python script, I was =
seeing
> 81 inserts/second and 30 updates/seconds for 100 kB documents. This is =
on a
> single 7200 rpm SATA drive. A write performance of 3 MB per second =
seems
> low.
>=20
> In brief, there are two things I can learn more about: Is is expected =
for
> the Document Update API to be slower? Two, does this kind of write and
> update performance look right?
>=20
> This is a Ubunut 10.01 system with
>=20
> 4GB RAM
> 900GB 7200 rpm SATA drive
> CouchDB 1.0.2 compiled using https://github.com/couchone/build-couchdb
>=20
> And the stock CouchDB 0.10.0 from Ubuntu has similar performance as =
well.