> Leyne, Sean wrote:
> >> >So, it is possible that you can get significantly faster disk IO
> >> >from such a solution. At last year's MS NA TechEd, several vendors
> >> >where showing Windows 2012 Storage Clusters which had Disk IO
> >> >throughput of 5*GBytes* per second!!!
> > A clarification/correction.
> >
> > The actual numbers were
> >
> > 512KB block size = 5792 MB/sec or 11,565 IO/s
> > 8KB block size = 2683 MB/sec or ***343,388 IO/s***
>
> I get the distributed disk structure thing. I'm using that already locally, but my
> network pipes can't cope with maximum rates :)
>
> But if it's serving 1000 different 'sites' in parallel ?

The "cloud" is not a single disk cluster or CPU cluster, it can be scaled to hundreds of clusters.

Not all of the 1000 sites would be necessarily by on 1 cluster.

It is all about a mixing-and-matching cloud instances, in a cluster, based on their performance profile...

Further, if you look at Amazon EC2, they offer both Virtual and Physical systems. So, you would choose you the instance/setup you need.

> My own sites are all cached
> across machines and are all generally in memory on the machines they are
> running from so no disk traffic is needed? I can't imagine that all of the sites
> will be 'live' on the cloud, so need to be loaded every time they are
> accessed?
> Surely the trade-off is between time it takes to reload an application and
> data to a location that can run it against simply accessing a real machine that
> has it available locally?

Don't quite follow your questions.

In a cloud, if you activate an virtual instance it runs until you (the instance owner) stop it. The instance is not suspended, so there is no re-loading from disk.

Remember that most instances are priced based on specific performance definition and usage expectations. So, you are paying for availability.


Sean