Delay Tolerant Bulk Data

Transfers on the Internet

OffenseMarcel Flores, Alok Rakkhit

The Proposition

•Present a method of doing large transfers over the internet

•Eliminate high link costs

•Eliminate need for couriers

•Don’t damage QoS for interactive users

•Seems like a reasonable goal...

The Solutions

•E2E-sched

•Use intelligent scheduling to transfer in the off-hours

•SnF

•Store and forward within the TR

So how does it look?

•With their straightforward analysis...it seems to work

•Report that it beats the cost of couriers

•Pretty graphs

•SnF works well in places where E2E-sched has problems...

...

•Well. Yeah.

•The whole idea is pretty obvious

•If you transfer when there are no others on, you can transfer more, faster

•Working around time zones is not complex either...

Assumptions

•Assume each direction of traffic charged independently

•No delay in TR

•Always possible to have in-TR storage

•How reasonable are these? Any real data?

Some fundamentals

•E2E-sched

•Has clear limitations (>5 time zone difference results in poor performance)

•SnF

•Are resources always available?

•Does the placement affect the performance?

The Experiment

•Wait what experiment?

•They just used these formulas to compute the available bandwidth!

•They didn’t actually transfer any data.

•How do they know it will work that well!

Other Comparisons

•Why is this better than some of the specialized protocols

•Even though they have different design goals, still worth a side by side

•QBSS - QBone scavenger

Applicability?

•Who can use this?

•Fine tuning likely depends on special information

•ISPs probably don’t want to share that information

•Could result in poorly tuned systems

Robust?

•Minute changes in ISP policy could do significant harm

•Combination percentile/volume pricing

•Other traffic shaping

•Increased Traffic?

•If this becomes popular

...but again!

•We don’t even know!

•They never ran any real tests!

•Not even in controlled environments

•It’s really only half the system