Re: Parameters which affect Performance of MySQL Cluster - Kenji HIROHAMA - com.mysql.lists.cluster

9 messages in com.mysql.lists.clusterRe: Parameters which affect Performan...

From	Sent On	Attachments
Kenji HIROHAMA	11 Dec 2005 23:13
Martin Skold	12 Dec 2005 03:15
Kenji HIROHAMA	12 Dec 2005 15:15
Adam Dixon	12 Dec 2005 16:25
Martin Skold	13 Dec 2005 02:23
Kenji HIROHAMA	13 Dec 2005 17:42
Martin Skold	14 Dec 2005 00:13
Kenji HIROHAMA	15 Dec 2005 01:52
Martin Skold	15 Dec 2005 04:24

Subject:	Re: Parameters which affect Performance of MySQL Cluster
From:	Kenji HIROHAMA (kenj...@gmail.com)
Date:	12/13/2005 05:42:31 PM
List:	com.mysql.lists.cluster

Let me make sure; # of SQL Node: 1 # of Data Node: 2 # of Replicas:2 engine-condition-pushdown=1

Under the above env., two Data Node have the same data, but data is partitioned two parts, F1 and F2. Let's say, Data Node A has primary replicas of F1 and secondary replicas of F2, and Data Node B has secondary replicas of F1 and primary replicas of F2.

With engine-condition-pushdonw=0, full table scan is executed at SQL Node after gathering whole row from Data Node A, or Data Node B. (in which TC response in this turn.)

With engine-condition-pushdown=1, full table scan is executed at Data Node A or Data Node B, in which TC responses in this turn.

Is my understanding correct?

BTW, the result of "engine-condition-pushdonw=1" didn't get better than I had expected.

Regards, Kenji

On 12/13/05, Martin Skold <Mart...@mysql.com> wrote:

Yes, conditions are pushed to all data nodes which contain fragments of the table being scanned. The scan is executed in parallel on all these nodes and conditions are applied to each tuple found to filter out what is to be sent back to the SQL node (this is why we sometimes call these "scan filters").

BR -- Martin

Kenji HIROHAMA wrote: Hi,

Does "pushdown" work even if # of Data Node is two and # of Replicas is two?

Regards, Kenji

On 12/12/05, Martin Skold <Mart...@mysql.com> wrote:

Hi!

If you are using 5.0, then setting --engine-condition-pushdown=on might greatly optimize:

full table scan with SELECT LIKE '%xxx%'

(if table is large)

BR -- Martin

Kenji HIROHAMA wrote:

Hi,

Is there any MySQL Cluster parameters which affect performance except --ndb-force-send, --ndb-use-exact-count, and ndb-optimized-node-selection?

With the TPC-W-like-benchmark tool (http://sourceforge.net/projects/osdldbt), TPS is not so good and server resources seem not to be used effectivelly, that is, CPU of Data Node/SQL Node is not utilized more than 70%..

The characteristics of the benchmark are the following;

-one transaction consists from several SQL queries -autocommit=off -full table scan with SELECT LIKE '%xxx%' -the number of connection is 100

I would very appreciate it if you could give any tiny piece of advice/comments/etc.

Regards,

-- kenj...@nospam.gmail.com Kenji Hirohama

-- Martin Sköld, Software Engineer MySQL AB, www.mysql.com Office: +46 (0)730 31 26 21