Some thoughts on what we (from NCSA's Virtual Director and NLANR/DAST groups)
hope to do re network visualization.
3-D interactive graphics
We do 3-D interactive graphics, and we believe that some relevant
visual metaphors work well in 3-D, so our tools will have a 3-D component.
Thinking of vBNS (and likely I2) backbone and institutional data
- scores/hundreds of nodes: can show all in one view (or even NxN matrix)
- fairly homogeneously collected data
what data?
- usual stuff: bandwidth, capacity, delay, loss, etc. by piece
(link, switch, router, AS)
- derived data
- fluctuations of above sampled over {interval} during {span}
- aggregates. what kinds of aggregates are most useful?
- animation over time
superpose general network view with single-app perf measures
- if our app is slow today, can we guess why?
- what kinds of tuning might help?
show comparison with normal behavior, as well as absolute measures
- what's "normal"?
- stats of this piece (link, AS->AS, ...) during
or of this group of pieces ("all OC12 links")
- what's "compare"?
- any cleverer measure than "std devs from mean" or "beyond Nth centile"?
- "abnormally abnormal"
- want to treat results of comparisons just like other derived data,
usable for further processing just as raw data can be
issues
- data indexing
- derived data is a lot messier than the raw data
- raw data is bulky -> processing expensive -> don't duplicate effort
- but how to remember what's already computed?
- even raw data isn't really homogeneous
- network topologies change
- want to show net history across changes
- recordkeeping styles change too, there'll be lost data, etc.
- historical vs. live network views
- are these so different that we'd want different tools to show them?
- hope not. certainly want to compare live stuff with past norms.
- but the way of importing data may be very different.
virtual reality
- some of us are really chartered to do (collaborative) VR, so we'd want
to implement (at least some) of our tools for virtual environments
- but they'd be useless if not also usable as desktop tools (who has a CAVE?)
- beyond making really slick demos,
can it help network
engineers or networked-application developers to be immersed in
the network view?
- likewise for collaboration:
any point in having widely-separated
people see and talk about a shared view of network statistics?
Stuart Levy, slevy@ncsa.uiuc.edu