Re: RAC versus simplistic load-balancing of replicated databases

Kevin Murphy wrote:

> Q1) would scalability best be achieved by:
>
> A. Real Application Clustering, or
> B. Some sort of load balancing across multiple machines each running
> an independent copy of the database?
>
> (UPenn has an Oracle site license; otherwise we would be using MySQL
> or PostgreSQL. )
>
> Q2) What would be the best means of doing the replication in Q1.B
> above?
>

Not quite sure why you are so worried about the database portion for a system this small ......... 2-5 GB data???? - you can't even buy disks that small any more! you buy memory chips that size! This whole thing would economically fit in memory on ram disk and you are concerned about scalability at the DB level?

As others have indicated, appropriate tuning is the first step (esp. in a proper load generating proof/benchmark environment).

And as Sybrand points out, the tuning at the app side needs to be reviewed carefully - likely more carefully than the DB side. There are a lot of brain-dead design patterns that look great at first blush, but are anti-scalable. (IMHO, most of these are really the work of advanced prototypers, many of whom do not know how to spell database.)

If you really want to do replication for this, you might want to review the documentation on transportable tablespace to see whether it fits your scenario.

>
> Q4) In general would you run the webserver instances on the database
> machines, or have the webservers run on separate machines?

Unless the machine is so oversized that no tuning is required at all, it is a reasonable idea to avoid mixed loads such as database and http services on the same machne. If you are trying to save hardware money, it is a stupid idea to oversize the machine to avoid the need to tune the subsystems properly. (Translation - separate machines).

> In either case, the load-balancing would essentially happen at the
> HTTP request level, since the webservers will keep database
> connections open. In the latter case (separate webservers), each
> webserver would effectively be wired to one database server, but if we
> put an additional load balancer in front of the database machines,
> that would at least allow all the webserver machines to obtain DB
> connections if one of the DB machines goes down. Is that correct?
> Some web/application servers might support software failure of DB
> connections also, I guess.

You might want to look at Oracle's applciation server, specifically the web cache part. Not sure, but I thought the documentation actually talked about load balancers and possible built-in balancing.

Also I recommend reading up on TAF (transparent application failover). And OTN (Oracle TechNet at http://otn.oracle.com) has an excellent set of pages on high availability worth (re-)reading.

> Q5) Are there any other questions I should be asking?

"Which documentation or parts are most applicable?" "Where do I find the relevant documentation?" (you diid ask! <g>)

> Thanks,
> Kevin Murphy
Received on Tue Jun 03 2003 - 23:03:44 CDT