State assumptions
Not mobile code
Not dynamic path creation
Not doing service discovery
Who owns clusters?
Assumption valid for large set of applications
Expected contribs before outline
Harvest -- related work
Related work -- Berkeley intensive
End of related work -- why rel work don't do what you're doing
Next generation portals -- composing services; MS won't do it,
Yahoo might
State assumptions about business model
Criterion for feasibility
Throw in other loss characteristics -- uniform...
What are the architectural issues -- before design
Routing -- lots of details...
Problem you're solving -- service instance choice -- motivate this
Choice of test sites...
Berkeley+UNSW -- too wide
Worst-case?
Developing prototypes
Going to be much more regionalized...
Who owns the services?
Be very clear about assumptions...
Better estimation of the size of overlay.  Scalability?
Cumulative distbn?
Why 4000/5000 nodes?
What is the end-product?  A little vague?
What are you actually going to build?
Emulation on millennium?
What is the product that is come out?
What are the pieces of the architecture?  How do you know when you're done?
Flow of info between pieces - business model or operational model
Characterize end services.. detailed characterization
Were'nt enough real examples
Music playback -- what is quick recovery?
Internet jitter -- vague -- give a reference
Cite -- and biblio
Backup slides...
What are the kinds of optimization metrics?
Why rel work does'nt look at this?  Have'nt looked at geographically
dist indep service providers -- bring that out in the title itself
Rel work -- IDMaps
How do you manage state?
How the big picture related to service providers?
Dichotomy between bigger picture and details
What are you trying to show in the evaluation?  What are the impact of the
results
Limit the area you cover -- do one thing in depth
Research methodology -- too many metrics; some notion of direction
Too long on the UDP stuff -- small expt -- less details -- ground it better
Evaluation -- should'nt spend too much time on this as well.
Rel work -- don't just throw in stuff -- say what's good about them
USITS paper -- TRIADS stuff
Collapse the related work