Imri Zvik wrote:
Hi,
You're very much correct about the nodes & HA - but that wasn't the
question - The question was if 3 nodes are a must -
Sorry I didn't see the question :-)
But except for testing I would not recommend using all the nodes on a single host.
 and the answer is no
- you can technically use 1 physical node. The fact it doesn't provide
any redundancy is not MySQL specific, but true to any cluster of any
kind :)

You are also correct about the indexes. As far as I can remember there
is a script that analyzes existing databases and reports the amount of
RAM needed to run MySQL cluster (it actually outputs the configuration
directives needed, but you can easily calculate the amounts from them).
yep ndb_size.pl

Bottom line, the MySQL cluster COULD be sufficient for Amos's needs, but
further data must be gathered before a decision could be made.




-----Original Message-----
From: Cyril SCETBON [mailto:[EMAIL PROTECTED]
Sent: Tuesday, November 20, 2007 12:13 PM
To: Imri Zvik
Cc: Yonah Russ; [email protected]
Subject: Re: MySQL Cluster for HA? What about PostgresQL PGCluster?



Imri Zvik wrote:
Hi,

The MySQL cluster can run perfectly on one physical node.

The 3 nodes you're referring to are at least 1 management node, and
at
least 1 API ("mysqld") node, and at least 1 data/storage ("ndbd")
node.
They can all reside on the same physical node.
yeah, but you lose the high availability if the host crashes.
The main issue with NDB version 5.0 (from my point of view) is that
it
must load all data into memory at runtime, so if your database is
larger
than your amount of RAM, you better find some other solution, or
split
it to enough replicas (data nodes) so that each chunk would fit.
This problem is going to be solved with 5.1,
just for the data blocks cause indexes are still in memory.
but it is not even near
being production-ready (at least last time I've checked).
yes, but the current version is an RC (release candidate) which
announces a release in the next 2 months.

-----Original Message-----
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf
Of
Yonah Russ
Sent: Tuesday, November 20, 2007 6:53 AM
To: Amos Shapira
Cc: [email protected]
Subject: Re: MySQL Cluster for HA? What about PostgresQL PGCluster?

Hi,

AFAIK MySQL cluster requires using the NDB engine which is not 100%
compatible with MYISAM, etc. The more recent a version you have the
better off you will be but there was one point where autoincrement
wasn't even supported. Also from what I know, a NDB cluster needs
at
least three nodes.

That doesn't mean you can't have an active-active database. If you
application is ready for it, you can set up MySQL in a Master
Master
configuration (aka multimaster). For your application to be
compatible
you basically need to use only autoincrement primary keys for all
your
tables but you should read the whole Megilah - here is a good
article:
http://www.onlamp.com/pub/a/onlamp/2006/04/20/advanced-mysql-
replication.html

Google has also released some code for multimaster configurations
and
they have some docs also:
http://code.google.com/p/mysql-master-master/

Regarding Postgres- I can't recommend anything specific but I am
looking into pgpool and pgpool II for a new project.

Hope that helps,
Yonah


On Nov 20, 2007 2:51 AM, Amos Shapira <[EMAIL PROTECTED]>
wrote:

        Hello,

        We are looking at enhancing one of our servers, which runs
CentOS
5 on
        amd64, for HA using Linux Virtual Servers (LVS). The server runs
a web
        site using MySQL 5.0.22 and of course we'll need to replicate
that on
        the second server as well.

        Is MySQL Cluster the right solution for us?

        At least for now, we think we can do with a master-slave
solution
        (i.e. only one server serving while the other is in "hot
standby"),
        but if it's possible to take advantage of both servers at the
same
        time and double our capacity then it would be very useful too.

        I also have a small web service using PostgresQL (8.1.9, from
CentOS
        packages) on the same server which will also have to be
replicated.
        What do people usually use to do this? Is PGCluster a practical
        option?

        Thanks,

        --Amos


=================================================================
        To unsubscribe, send mail to [EMAIL PROTECTED]
<mailto:[EMAIL PROTECTED]>  with
        the word "unsubscribe" in the message body, e.g., run the
command
        echo unsubscribe | mail [EMAIL PROTECTED]



===============================================================
To unsubscribe, send mail to [EMAIL PROTECTED] with
the word "unsubscribe" in the message body, e.g., run the command
echo unsubscribe | mail [EMAIL PROTECTED]






=================================================================
To unsubscribe, send mail to [EMAIL PROTECTED] with
the word "unsubscribe" in the message body, e.g., run the command
echo unsubscribe | mail [EMAIL PROTECTED]

Reply via email to