Also, is there any plans to inplement distributed joins for raid b-2? Perhaps through a query parser in the controller, which detects a join-type statement, then contacts a global catalog db/table on the db servers that contains a list of the nodes that contain the table requested?
-Tony --------------------------- Manager, IT Operations Format Dynamics, Inc. 303-573-1800x27 [EMAIL PROTECTED] http://www.formatdynamics.com -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Anthony J. Biacco Sent: Tuesday, February 26, 2008 9:54 AM To: Sequoia general mailing list Subject: RE: [Sequoia] still being developed? (questions) Thanx you for your answers Emmanuel. My hesitation to use b-0 is the same as you pointed out, the distributed joins. And being that it is the same for b-2 also, i'm not sure just using b-1 (without a SAN) would work for us, as we plan to scale to a larger database. probably 2TB+. Even being that disk space is cheap these days, i'm not sure if doing 2TB in each node with a small number of disk spindles is the right way to go. Especially, as i increase the number of nodes and transactions. I may end of going with the paralleldb option, san, and see how MySQL w/GFS performs with a single storage instance. -Tony --------------------------- Manager, IT Operations Format Dynamics, Inc. 303-573-1800x27 [EMAIL PROTECTED] http://www.formatdynamics.com -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Emmanuel Cecchet Sent: Tuesday, February 26, 2008 4:22 AM To: Sequoia general mailing list Subject: Re: [Sequoia] still being developed? (questions) Hi Anthony, > 1. I understand that using sequoia kind of assumes a shared-nothing disk > architecture. But is it possible to use it with a SAN instead of RAIDed? > If you use a SAN it means that your database supports shared-disk clustering (like Oracle RAC) and already deals with the proper locking and replication mechanism. > Can you even specify no RAID level? If your database readily supports replication (e.g. using a SAN), then you can use the ParallelDB load balancer that just forwards requests to your database cluster. In this case, Sequoia brings you load balancing, transparent failover and eventuelly request caching. > If so, does sequoia then have > nothing to do with this and all the reading/writing would just fall back > to MySQL's standards and locking along with whatever filesystem i'd be > using (probably GFS)? Would this even be possible or plausible? > Just to make sure that this is clear, Sequoia just relies on the capabilities of the underlying databases. So if you try to use a SAN with a database that has no support for it, Sequoia will not help in any way. > 2. In trying to understand the RAIDed architecture in regards to > sequoia, i'm not sure i understand something about the b-2 level. Does > sequoia determine which tables go on which servers? Or would I determine > that schema? What happens when an extra database server is introduced > into the mix? Does the schema re-partition the data automatically? Does > this need to be done manually? > In RAIDb-2, Sequoia will fetch the schema that is available on each replica and proceed from there. If you want to force only specific tables on each node, this can be specified manually in the virtual database configuration file. The same thing applies when a new table needs to be created, you can specify policies to create that table on a specific set of nodes or choose dynamically nodes among a pool to select the less loaded ones. > 3. For nested raid (b-0-1/b-1-0), is this a common config, and if so, > can someone comment on how well this works in the real world, in regards > to performance and ease of recovery? > No this is not a common config and RAIDb-0 is rarely used because it does not support distributed joins. If table A is on node 1 and table B on node 2 and you try something like SELECT * FROM A,B you will get an error that no node has both tables and the query cannot be executed. > 4. In regards to the controller, where is this recommended to be placed? > on the app servers (tomcat), or the database servers? > There is no single answer to that. It depends how many app servers or DBs you have. Sometimes you can collocate everything Tomcat/Sequoia/DB and sometimes you use even dedicated machines for Sequoia controllers. So all combinations are possible. Thanks for your interest in Sequoia, Emmanuel -- Emmanuel Cecchet - Research scientist EPFL - LABOS/DSLAB - IN.N 317 Phone: +41-21-693-7558 _______________________________________________ Sequoia mailing list [email protected] https://forge.continuent.org/mailman/listinfo/sequoia _______________________________________________ Sequoia mailing list [email protected] https://forge.continuent.org/mailman/listinfo/sequoia _______________________________________________ Sequoia mailing list [email protected] https://forge.continuent.org/mailman/listinfo/sequoia
