Hot off the press! I was just at the launch of the new Exadata Database Machine Version 2, made by Sun and Oracle. This combined OLTP and DW database machine delivers an astonishing One Million Random IOPS per cabinet and it scales to 8 cabinets! Larry talked about the problems of scaling a traditional non-Oracle database and adding capacity on demand to a database on the Cloud.
Assume you have a database running on a single server and you max out your server. What are your options? Well, if you are running a traditional database, you need to find a bigger box and then migrate your database to the newer server. You would need to size your new box with spare (potentially wasted headroom) capacity for future growth. If your database is partitioned across many servers, then you would need to add a new box, migrate data, repartition and try again to see where the hot spots are. With virtualization you can reduce some of the overhead but unless you are using an efficient Para-virtualized kernel (think Oracle VM based on Xen) you could still have issues scaling the database I/O.
If you are running on a Cloud, how do you know who is going to use the system? or how big a server you can deliver? Will it be big enough? How do you scale the I/O, what happens when you run out of HBA slots? Sizing and capacity planning on the Cloud are big headaches and there is a potential to be very inefficient.
The new Exadata Database Machine Version 2 gets round both of these issues. As the underlying Oracle Real Application Clusters (RAC) database is a shared data system (as opposed to a shared nothing) it is automatically shared across all available server nodes. You can start small, with a half rack and then drop in boxes to automatically scale to 8 cabinets. This database appliance really simplifies the installation, cost, capacity planning, integration, support and management of some significant Cloud infrastructure issues. The best of SUN and Oracle that can finally deliver a Cloud-in-a-box solution to the enterprise.
Can’t wait to get my paws on one of these babies…. now I just have to find a customer that wants to run a Cloud in the 8 Million random IOPS range…. any takers?
