James,
In the proposed architecture, you could directly attach the backends to
the RAIDb-1 controllers and have a full RAIDb-1 configuration.
So do you mean avoid a nested configuration, and just run RAIDb-1 across
all four nodes (4*replication) with two controllers? This might actually
be a viable solution.
Yes, you should avoid nested solutions because they make the management
much more complex and add an extra hop that will increase latency. It is
a good tradeoff to have 2 controllers with 2 databases each and run
RAIDb-1 across the entire cluster.
This should be similar to the distributed RAIDb-1 demo configuration
shipped with Sequoia.
You can
also have the top controllers use RAIDb-2 but you should be careful
about table distribution to make sure that no distributed join is
required (not supported)
As far as I know, there is no requirement for distributed joins in
DSpace.
Just to be sure that this Sequoia limitation is clear, I will give you
an example. If you decide to have table A only on node1 and table B only
on node2, Sequoia will not be able to execute queries like SELECT ...
FROM A,B WHERE ... or UPDATE A SET x=x+1 WHERE A.id=B.id.
Absolutely; in DSpace, the vast majority of activity will be reads.
Inserts, updates, and deletes will be (for the most part) few and far
between.
So it is safe to use RAIDb-1 in that case, you'll get the best
performance/availability for your application.
Don't hesitate to post on the mailing list if you need any help.
Actually, I did a few months ago and got no help :(
http://www.mail-archive.com/[email protected]/msg01129.html
Sorry about that, don't hesitate to insist if we forgot to answer,
sometimes we are overwhelmed with other activities and overlook some
mails on the list.
(First google hit for "raidb nested configuration"!)
Aside from the issues I encountered with the nested configuration, I hit
some pretty serious problems with (I think) the way that the postgres
JDBC driver dealt with UNICODE strings from the database. I'll try to
recreate them again, and I'll post to the list with the error. I
remember checking at the time, and it was a "known issue" with Sequoia +
PostgreSQL + UNICODE.
Ok, don't hesitate to keep us posted.
Thanks again for your feedback,
Emmanuel
--
Emmanuel Cecchet
Chief Scientific Officer, Continuent
Blog: http://emanux.blogspot.com/
Open source: http://www.continuent.org
Corporate: http://www.continuent.com
Skype: emmanuel_cecchet
Cell: +33 687 342 685
_______________________________________________
Sequoia mailing list
[email protected]
https://forge.continuent.org/mailman/listinfo/sequoia
