I might be missing something, but it seems to me that you are
comparing using CloudFront to trying to build your own CloudFront.
Building your own does not seem like it would be very easy or cost
effective.

Essentially, S3 is an NFS, innit? We use it that way.

What is the issue with CloudFront?

Cary

On Thu, Mar 1, 2012 at 2:36 PM, Kyle Banerjee <[log in to unmask]> wrote:
> Howdy all,
>
> I have no experience with autoscaling or streaming, so I'm looking for
> thoughts that help me wrap my mind around how to implement it in a
> production setting.
>
> I have been asked to examine the possibility of providing a consortia level
> music reserves system using Variations (which I also have no experience
> with). The software would be maintained centrally, but each institution
> will manage its own collections, users, etc. Load is expected to vary
> considerably, varying from practically nothing to possibly hundreds of
> simultaneous streams at peak time. This strikes me as an excellent elastic
> application.
>
> This strikes me a good EC2 app and as far as I can tell, there are two
> basic ways to achieve the elasticity I'm looking for in that environment.
> The first is to store the files in S3 buckets and serve them via
> CloudFront. This strikes me as the preferred solution, but I don't yet know
> if I'll be able to get the user and staff clients to play well with this
> configuration. The second is to have a script monitor the service and spin
> up more instances when certain triggers are met and destroy them when
> demand drops. But if I do that, all instances need to be able to access the
> same live data. For DB data, that's a no brainer since I can just run a DB
> server. But how do you synchronize live files across instances since EBS
> volumes can only be accessed by one instance? Somehow, NFS strikes me an
> ugly way to deal with the problem. Actually even if EBS volumes could be
> attached to multiple instances, that solution would still suck as you could
> have multiple apps trying to access files at the same time.
>
> Obviously, I'm having trouble getting pointed in the right direction. I
> could punt and just order capacity to handle heavy use cases. But that's a
> copout, and figuring out autoscaling bandwidth and computing capacity just
> feels like one of those tools that's really handy to have in your bag of
> tricks. Any pointers would be appreciated.
>
> kyle
>
> --
> ----------------------------------------------------------
> Kyle Banerjee
> Digital Services Program Manager
> Orbis Cascade Alliance
> [log in to unmask] / 503.999.9787



-- 
Cary Gordon
The Cherry Hill Company
http://chillco.com