noisetor-01 was pushing 300-400 Mbps of traffic from 2012-02-15 through 2012-04-13. Since mid-April we've seen traffic decrease significantly; over the last week, our daily peak has been 260 Mbps (versus 450 Mbps in March) and our daily trough has been 100 Mbps (versus 300 Mbps in March).
The traffic levels dropped off in a smooth fashion over a 10 day period, April 15-25 if I read the graphs correctly.
There's no sign that anything is amiss on the node; CPU and memory usage are consistently lower, today, than they were in March -- as expected, given the lower utilization. The logs are silent.
I've been postprocessing torstatus's Tor_query_EXPORT.csv regularly, and it appears that the overall exit throughput of the network has gone up slightly over the last few months; some new significant nodes came online during 2012Q1 and may have affected the traffic distribution. However, while noisetor has seen a significant drop, other families of exit nodes have seen their bandwidth go up or stay steady.
Has there been a change in the routing algorithm, or any other network changes that might explain this drop?
Thanks, -andy
Thus spake Andy Isaacson (adi@hexapodia.org):
noisetor-01 was pushing 300-400 Mbps of traffic from 2012-02-15 through 2012-04-13. Since mid-April we've seen traffic decrease significantly; over the last week, our daily peak has been 260 Mbps (versus 450 Mbps in March) and our daily trough has been 100 Mbps (versus 300 Mbps in March).
The traffic levels dropped off in a smooth fashion over a 10 day period, April 15-25 if I read the graphs correctly.
Has there been a change in the routing algorithm, or any other network changes that might explain this drop?
The loss of non-TBB Torbutton users might explain a drop post Apr 20th, when their ability to move tabs around was borked by FF12: https://bugzilla.mozilla.org/show_bug.cgi?id=715885#c33
But otherwise, according to https://metrics.torproject.org/consensus-health.html, all 5 bw auths are voting, and I have not changed the algorithms.
On Wed, May 02, 2012 at 08:29:08PM -0700, Andy Isaacson wrote:
Has there been a change in the routing algorithm, or any other network changes that might explain this drop?
I opened https://trac.torproject.org/projects/tor/ticket/5755 for a related topic that I think will help answer questions like this (or at least narrow out possible explanations).
So long as https://metrics.torproject.org/performance.html#torperf is looking steadily good, I'm not too worried about individual fluctuations in relays -- there is currently at least slightly more capacity in the network than is needed to keep the torperf results at that level.
Another resource you might find interesting is http://freehaven.net/~arma/moria1-v3-status-votes which lets you look up the individual weighting votes for your relay from the most recent consensus.
In theory, if your capacity has stayed the same but your load has dropped, then the bandwidth authority measurements should show you as getting faster. I wonder what happened in practice. Last I checked, Mike didn't want to publish any intermediate statistics from the bandwidth authorities, because they were too voodooey.
--Roger
On Thu, May 03, 2012 at 01:09:30AM -0400, Roger Dingledine wrote:
I opened https://trac.torproject.org/projects/tor/ticket/5755 for a related topic that I think will help answer questions like this (or at least narrow out possible explanations).
[snip]
In theory, if your capacity has stayed the same but your load has dropped, then the bandwidth authority measurements should show you as getting faster. I wonder what happened in practice. Last I checked, Mike didn't want to publish any intermediate statistics from the bandwidth authorities, because they were too voodooey.
https://trac.torproject.org/projects/tor/attachment/ticket/5755/consensus-we...
It looks like the weight assigned to your relay, as a fraction of total weights, dropped in mid April.
So either there's even more capacity to go around (good outcome), or Mike's bandwidth authority scripts decided you should get less attention (not so good outcome).
Mike, perhaps you can look at the bwauth data files and take a guess?
--Roger
tor-relays@lists.torproject.org