On Fri, Oct 16, 2009 at 9:48 PM, Adrian Cole <adrian@jclouds.org> wrote:
I postulate that regardless of java, python, curl, (name whatever you like), clients will be more efficient and capable on retrying transient failures if they are not required to store entities along with their requests. I'll challenge anyone to write a test in any language showing parsing entities as faster, more resilient to retries, or more scalable then 0-length requests using matrix or headers for things like this=true or that=2000 :)
Agreed. Another thing to remember is that by diving in the main stream with the entities themselves you're taking on a good deal more work and forcing your clients to do the same. Not only do you have to define some domain-specific language for the task at hand but your clients then have to both parse and understand it which creates a very significant barrier to entry compared to using existing open standards like OVF. Note that this isn't so much as a change in tack for me as a separation of duties - I'm still far from convinced that OVF is great for the industry (I see it becoming another MS Word document format if we're not careful) but standardising a plain text format like VMX is something we can tackle separately later if need be. Sam