Opened 6 years ago

Last modified 17 months ago

#7357 new enhancement

Collect more statistics to better understand simulations

Reported by: robgjansen Owned by:
Priority: Medium Milestone: Tor: unspecified
Component: Core Tor/Tor Version:
Severity: Normal Keywords: performance, simulation, statistics, tor-relay, tor-client research-program
Cc: arma, karsten, robgjansen, teor Actual Points:
Parent ID: Points: parent
Reviewer: Sponsor:

Description (last modified by robgjansen)

Our simulations often lead to observations that indicate a lack of understanding of the results. This is because the information currently gathered is very high level: client download times and relay load.

We should start collecting more information inside Tor that will allow a more complete analysis of a simulation to be able to say things about congestion, bottlenecks, and which nodes are responsible for certain behaviors. This should improve our ability to reason about future design changes and how they affect Tor's various statistical properties.

This standardize stats across simulations/real experiments/tor metrics.

Child Tickets

TicketTypeStatusOwnerSummary
#6781enhancementnewMake cell statistics useful in simulations again
#7134projectneeds_informationAdd statistics on time spent on crypto operations
#7358tasknewDecide on list of stats to collect
#7359enhancementclosedDesign/implement method for collecting/reporting statistics
#7360tasknewDocument process of collection and extraction, formatting, etc.
#7382enhancementnewDesign/implement technique for measuring/collecting inter-relay network latency

Change History (7)

comment:1 Changed 6 years ago by arma

Agree.

Is there a list of what things you want to know, or is that step zero?

comment:2 in reply to:  1 Changed 6 years ago by robgjansen

Replying to arma:

Agree.

Is there a list of what things you want to know, or is that step zero?

You're too fast, i'm still creating the tickets ;) See #7358.

comment:3 Changed 6 years ago by karsten

I like the idea of adding more statistics. But just to be sure, this is not about adding any aggregate statistics that relays report to the directory authorities that we then collect and make public? It's just about making clients and relays with the TestingTorNetwork or some other safeguard option set report data about their operation to their control port, right? Phrased differently, there will be no need to touch dir-spec.txt, just control-spec.txt?

comment:4 in reply to:  3 Changed 6 years ago by robgjansen

Replying to karsten:

I like the idea of adding more statistics. But just to be sure, this is not about adding any aggregate statistics that relays report to the directory authorities that we then collect and make public? It's just about making clients and relays with the TestingTorNetwork or some other safeguard option set report data about their operation to their control port, right? Phrased differently, there will be no need to touch dir-spec.txt, just control-spec.txt?

I would say yes for now. Though, I can certainly imaging that some stats would be useful to aggregate and report. Having [more] real live network stats would improve our understanding of how realistic our simulated networks are. But yes, that transition, if it ever happens, can happen separately.

comment:5 Changed 6 years ago by robgjansen

Milestone: Tor: 0.2.4.x-finalTor: unspecified

comment:6 Changed 6 years ago by robgjansen

Description: modified (diff)

comment:7 Changed 17 months ago by nickm

Cc: teor added
Keywords: research-program added
Points: parent
Severity: Normal
Note: See TracTickets for help on using tickets.