Opened 9 years ago

Closed 9 years ago

Last modified 8 years ago

#2534 closed task (fixed)

get torperf output up on metrics website

Reported by: arma Owned by: karsten
Priority: Medium Milestone: Deliverable-May2011
Component: Metrics/CollecTor Version:
Severity: Keywords:
Cc: Actual Points:
Parent ID: Points:
Reviewer: Sponsor:

Description

Karsten, please revise this text to enumerate what torperf data there is that isn't getting recorded (or rather, that wasn't getting recorded as of Dec 1), and a plan for getting it archived on the metrics page.

For example,
https://metrics.torproject.org/data.html#performance

Child Tickets

Change History (5)

comment:1 Changed 9 years ago by karsten

There's no Torperf data that isn't getting recorded, as far as I know. Here's the list of Torperf data that we have on the metrics website:

  • Timing data from moria (July 3, 2009), ferrinii/torperf (Oct 24, 2009), and siv (March 8, 2010)
  • Path data from moria (January 26, 2011), ferrinii/torperf (January 26, 2011), and siv (January 31, 2011)
  • Timing and path data from ferrinii/torperf with custom guard selections (February 1, 2011)

We only have graphs for the timing data from our three main Torperf setups. But we record the data for the other Torperf runs, too, so we can graph them whenever we want.

comment:2 Changed 9 years ago by karsten

Status: newassigned

Roger: Is your question answered by my comment? If yes, please close the ticket or tell me to do so. If not, what is needed to close this ticket?

comment:3 Changed 9 years ago by karsten

Resolution: fixed
Status: assignedclosed

Roger says this is fine to close. Closing.

comment:4 Changed 9 years ago by arma

Looks good. We should make a habit of collecting output data from any torperfs we run, since we never know when we'll realize we wanted some old data set.

comment:5 Changed 8 years ago by karsten

In theory, I agree that we should keep all output data of all Torperf runs. But in practice I cannot guarantee this. We're heavily changing Torperf's output data format these days and are running the 12 Torperfs with changed guard node selections in various configurations. Some of these experiments run for a few days only. Documenting everything and providing the data for others to review would require some additional effort that I'm trying to avoid for this experiment. So, here's what I think what we can guarantee and what we shouldn't guarantee:

  • Right now, we keep the performance output data of the 3x3 Torperf runs on ferrinii, moria, and siv.
  • Once we have improved Torperf enough to output the path data we want, we can also guarantee to keep the path data of the 3x3 Torperf runs. We already provide these data, but we might break the output data while doing so.
  • The 12 additional Torperf runs on ferrinii are only an experiment that we're stopping and restarting periodically to improve Torperf's output data. We cannot guarantee that we keep all data for every instance of the experiment.
Note: See TracTickets for help on using tickets.