Stephen Hahn wrote:
>   3.  Akamai isn't the only CDN in the world.  It's not even the only
>       one Sun does business with--and all of them are interested in
>       doing compute-in-the-cloud.  Assertions about GET versus POST for
>       CDNs are short-term assertions.  More importantly, there is a
>       distinct pricing calculation to be made between directly-served
>       and CDN-served data that not every pkg(5) depot operator will wish
>       to make.

I think you're saying there that many different kinds of distribution 
mechanisms need to be supported.  I would never say that we should 
design *for* Akamai, because next year we might well use a different 
download provider.  However, Akamai is an example of a major download 
provider, and one that we *do* use, and so any solution should be 
designed to be compatible with it, and with any other mechanism that we 
can reasonably envision.

To me, that suggests that the protocol needs to be as simple and 
standard as possible, so that it does not have to be reimplemented for 
each distribution mechanism.  That in turn translates to "behave like a 
completely static file-oriented web server".

(Perhaps some people will respond with "but it's all HTTP, and that's a 
standard!".  That's an incomplete understanding.  There are layers upon 
layers, and here there's another protocol layer on top of HTTP:  the 
layer that defines the semantics of GET and POST operations and the URLs 
that they operate on.  If you all understood that, sorry for the 
distraction.)
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss

Reply via email to