Hello. I am attempting to set up a private cloud for hosting a database server cluster.
I have successfully deployed multiple VM's and am now looking at making the setup as efficient as possible.

Harddisk I/O is the major bottleneck for my designs, especially on any persistent drives I might need.
I am also looking into load balancing and task prioritization within the cloud.

I have a couple of questions regarding Eucalyptus architecture / features:

1. Am I right in my understanding that all EBS drives within a cluster reside ON the SC? i.e. all read/writes to/from EBS drives within instances cause network traffic and disk r/w on the SC. Is there a way to designate disk space from NC's to the pool of available space for volumes? Having X instances all performing tasks not only across the network but all onto one disk is slow. If anyone has any tricks for improving the performance of EBS drives I would be interested in knowing.

2. I'm thinking this one is a long shot, but is there a way to dynamically change the number of cores (maybe even RAM) available to an instance? If I have an instance that for the most part requires only a single core but occassionally takes high priority and requires more resources (all available on the same NC of course) without interrupting the instance (no reboot or redeployment), is there a way to make this happen?

3. On the node controllers themselves, does using a RAID0 setup for the physical machine make much difference to the instance disk operations? Has anyone performed any tests? I would think that using VirtIO Disk would produce a noticeable difference but would be interested in any research that anyone has done.