Opened 7 years ago

Closed 7 years ago

#5333 closed task (fixed)

Make simulator authors aware of "clients have different connectivity" wishlist item

Reported by: arma Owned by:
Priority: Medium Milestone:
Component: Metrics/Analysis Version:
Severity: Keywords:
Cc: robgjansen, kevin, iang@…, tschorsch@… Actual Points:
Parent ID: #4506 Points:
Reviewer: Sponsor:

Description

In #4487 I want to understand how n23 behaves when the client's first hop is flaky.

But all the Experimentor and Shadow graphs I've seen so far look like the clients all have the same connectivity, and the graphs are averages of client results.

I guess we could run the experiments where *every* client has a flaky first hop. But that wouldn't capture the case where most clients are hammering the network, and our poor client can't hammer the network because his connectivity isn't good enough.

Can the two simulators capture different client bandwidth/connectivity properties? How hard would that be to add?

(I ask because I hope to ask for simulation results early next week once we've resolved #4488.)

Child Tickets

Change History (5)

comment:1 Changed 7 years ago by Flo

Cc: robgjansen kevin iang@… tschorsch@… added; robgjansen kevin iang@… removed

comment:2 Changed 7 years ago by kevin

It would not be difficult for ExperimenTor to configure client links with arbitrary bandwidth and delay characteristics. I current sample asymmetric bandwidths from Ookla's Net Index data sets, to approximate broadband Internet connectivity (I believe that Shadow does something similar).

Can you elaborate more on what you mean by every client has a "flaky" first hop? What bandwidth / delays would reasonably approximate the behavior you want to investigate? Are there any data sets that might help us?

comment:3 Changed 7 years ago by robgjansen

In addition to bandwidth and delays, Shadow also models link loss rates. Would increasing the loss rate on links to certain clients achieve the "flakiness" you desire?

comment:4 in reply to:  3 Changed 7 years ago by robgjansen

Replying to robgjansen:

In addition to bandwidth and delays, Shadow also models link loss rates. Would increasing the loss rate on links to certain clients achieve the "flakiness" you desire?

I should mention that by loss rate I mean a probability L such that every packet that is sent over the link will have L chance of being dropped.

comment:5 Changed 7 years ago by robgjansen

Resolution: fixed
Status: newclosed

To reiterate, Shadow can model this. We can put clients in some far off land where they have high latency, low bandwidth, and high loss rates, or any combination thereof. And we can separate their performance from "the rest of the pack" fairly easily.

I want this ticket to be closed, so, lets define precisely what experiments we want in #4487 and carry them out there.

Note: See TracTickets for help on using tickets.