|
It also depends on what kind of database you're running. I know there's
issues with using GFS and Oracle on Linux. I bought this conversation up
before on the gfs list. Not being an Oracle administrator I didn't
understand the specifics of why it was bad, but I was told "don't do it
dude". Oracle is a big bulky mess of crap that makes a lot of money.
Smaller RDBMS's probably don't have the hang ups Oracle has with it's data
files. MySQL and Postgres probably work fine. GFS definitely provides a
pretty simple infrastructure for fair over on shared information.
-jeremy
>
>
> On Mon, 17 Jan 2000, Peter Koch wrote:
>
> > Hi,
> >
> > I'm pretty new to this group, and am currently experimenting with
> > moving to a linux high availability cluster. Fault tolerance is
> > essential, so I'm using a fake linux load balancer, but I also need to
> > make sure the database server is fault tolerant also.
> >
>
> Well, fault tolerance and high availability are different concepts. I
> think fault tolerance should be higher than high availability. If a
> service is down for seconds and a minutes and existing connections may be
> lost, then the service is up to accept new connections, we can say the
> service is highly available. Fault tolerance probably means that once the
> connection is accepted, it should be carried out, despite of partial
> hardware or software failure.
>
> > Can anyone give me some pointers as to how they are achieving database
> > access from the cluster, are you using a dedicated database server, or
> > does each server in the cluster have their own local database cache
> > that is kept synchronised with distributed transactions. If you are
> > using a dedicated database server(s) what is the impact to the cluster
> > since now local updates need to be done over the network.
> >
>
> If you just want to make a database system of 2 nodes highly available.
> You can have a look at http://www.linux-ha.org/ for heartbeat code, the
> primary database server and the backup can heartbeat, if the primary
> fails, the backup will take over the db service.
>
> If you want to build a highly scalable database cluster, there must be
> many works to do. I suggest that you use global storage like GFS to keep
> database files, and all database nodes can see those database files. And,
> there must be a Distributed Lock Manager to reconcile the conflict when
> multiple database nodes want to access the same file or data block
> concurrently. Finally, you can use LVS to group all the database nodes for
> a single highly scalable database system.
>
> Wensong
>
>
>
> ----------------------------------------------------------------------
> LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
> To unsubscribe, e-mail: lvs-users-unsubscribe@xxxxxxxxxxxxxxxxxxxxxx
> For additional commands, e-mail: lvs-users-help@xxxxxxxxxxxxxxxxxxxxxx
>
http://www.xxedgexx.com | jeremy@xxxxxxxxxxxx
---------------------------------------------
|
