Hi George,
I found this in my IRC backlog from November 7, 2014:
20:34 #tor-dev: < asn> karsten: how should one read these graphs https://metrics.torproject.org/bandwidth.html#bandwidth-flags ? 20:34 #tor-dev: < asn> karsten: can the gap between two lines be interpreted as available (unused) bw? 20:35 #tor-dev: < asn> karsten: so the gap between the yellow lines is the available guard bw? 20:35 #tor-dev: < asn> karsten: but there is not that much available exit bw, right? so it's not entirely accurate.
These are all fine questions.
First of all, the gap between two lines is *supposed* to be interpreted as available (unused) bandwidth. Advertised bandwidth is the number of bytes that all relay together were willing and capable to handle. (It's the minimum of the three bandwidth numbers contained in server descriptors.) Bandwidth history is the number of bytes that all relays together actually handled. (It's the sum of byte histories contained in extra-info descriptors.) So, the difference of the two should be the number of bytes that was left unused, in theory.
However, I can't say how accurate this comparison is. If I had the time to investigate, I'd start by downloading the latest descriptor archives from CollecTor and comparing advertised bandwidths and bandwidth histories per relay. If you or somebody else wants to do that I might be able to give them a hand.
All the best, Karsten
On Wed, Nov 12, 2014 at 09:30:47AM +0100, Karsten Loesing wrote:
Hi George,
I found this in my IRC backlog from November 7, 2014:
20:34 #tor-dev: < asn> karsten: how should one read these graphs https://metrics.torproject.org/bandwidth.html#bandwidth-flags ? 20:34 #tor-dev: < asn> karsten: can the gap between two lines be interpreted as available (unused) bw? 20:35 #tor-dev: < asn> karsten: so the gap between the yellow lines is the available guard bw? 20:35 #tor-dev: < asn> karsten: but there is not that much available exit bw, right? so it's not entirely accurate.
These are all fine questions.
First of all, the gap between two lines is *supposed* to be interpreted as available (unused) bandwidth. Advertised bandwidth is the number of bytes that all relay together were willing and capable to handle. (It's the minimum of the three bandwidth numbers contained in server descriptors.) Bandwidth history is the number of bytes that all relays together actually handled. (It's the sum of byte histories contained in extra-info descriptors.) So, the difference of the two should be the number of bytes that was left unused, in theory.
However, I can't say how accurate this comparison is. If I had the time to investigate, I'd start by downloading the latest descriptor archives from CollecTor and comparing advertised bandwidths and bandwidth histories per relay. If you or somebody else wants to do that I might be able to give them a hand.
At least on my exit (gurgle), there is in fact a lot of bandwidth going unused. It's configured for 12 MB (~ 100 Mbit), and has pushed those kinds of numbers in the past. But lately, it's been seeing more like 2-3 MB traffic (possibly because its "exit probability" value has gone down quite a bit?). There's probably something in the positive reinforcement feedback loop (your relay just by random chance happens to get fewer clients choosing it for a bit; it reports a lower bytes pushed value; it gets consequently assigned a lower consensus weight; fewer clients pick it in the future) that makes it hard for a relay to push the traffic it's capable of?
Are other exit relays hitting their resource caps, either in terms of CPU or bandwidth? Or does the whole Tor network just really have much more capacity than usage right now?
- Ian
-
Ian Goldberg -
Karsten Loesing