Dear list,
I would like to discuss hibernating and the consequences for the network. As someone running unmetered GBit systems i like to point out that there is a downside for the network when it comes to hibernating.
Usually a few days before the end of every month my systems are getting slammed with traffic / directory requests. I thought about that and came up with the theory that a lot of systems with traffic limitations are dropping out a few days before the end of the month. This means more pressure on the remaining systems in the network.
If the trend that more systems with limitations are participating increases we are going to see a serious imbalance in the network at some point. I know, poor unmetered systems ;-) I would like everybody to bear this in mind when it comes to the decision "Adjust the Rate or just open the gates and burn it as fast as possible".
I'm in the fortunate position to be able to tribute a nice amount of money / traffic, but even systems with unmetered traffic can just help until the bandwidth / hardware limits are reached. It would be awesome if i could conbince some of you to take a step back and take a moment to look at the bigger picture. I would like to provide a good service for everyone, even at the end of the month. That's getting harder the more systems are not present at the end of the month.
At this point i like to quote G. K. Chesterton:
“We are all in the same boat in a stormy sea, and we owe each other a terrible loyalty.”
Thank you for reading this and my best wishes for all of you.
On 02/01/2015 08:02 PM, Sebastian Urbach wrote:
Usually a few days before the end of every month my systems are getting slammed with traffic / directory requests. I thought about that and came up with the theory that a lot of systems with traffic limitations are dropping out a few days before the end of the month. This means more pressure on the remaining systems in the network.
I was wondering why the exit probability of my relay was tripled during the past few days and now is at the usual level. Your thoughts explains it well.
On Sun, Feb 1, 2015 at 2:29 PM, Toralf Förster toralf.foerster@gmx.de wrote:
On 02/01/2015 08:02 PM, Sebastian Urbach wrote:
Usually a few days before the end of every month my systems are getting slammed with traffic / directory requests. I thought about that and came up with the theory that a lot of systems with traffic limitations are dropping out a few days before the end of the month. This means more pressure on the remaining systems in the network.
I was wondering why the exit probability of my relay was tripled during the past few days and now is at the usual level. Your thoughts explains it well.
I wonder how hard it would be to have relays randomize the start point of their hibernation period, to stabilize the amount of available bandwidth over a 1-month interval...
zw
They already do so that you don't have all dropping at the same time.
--- GPG/PGP Fingerprint E129 722B A512 105C E8BE 4705 8046 EA48 2C82 1339 https://arlen.io/key On Feb 1, 2015 3:33 PM, "Zack Weinberg" zackw@cmu.edu wrote:
On Sun, Feb 1, 2015 at 2:29 PM, Toralf Förster toralf.foerster@gmx.de wrote:
On 02/01/2015 08:02 PM, Sebastian Urbach wrote:
Usually a few days before the end of every month my systems are getting slammed with traffic / directory requests. I thought about that and came up with the theory that a lot of systems with traffic limitations are dropping out a few days before the end of the month. This means more pressure on the remaining systems in the network.
I was wondering why the exit probability of my relay was tripled during
the
past few days and now is at the usual level. Your thoughts explains it well.
I wonder how hard it would be to have relays randomize the start point of their hibernation period, to stabilize the amount of available bandwidth over a 1-month interval...
zw _______________________________________________ tor-relays mailing list tor-relays@lists.torproject.org https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-relays
Zack Weinberg wrote:
I wonder how hard it would be to have relays randomize the start point of their hibernation period, to stabilize the amount of available bandwidth over a 1-month interval...
I run my relay with a daily limit using:
AccountingStart day HH:MM AccountingMax XXXGB
My relay currently runs for about 2/3 day and hibernates for the other 1/3 of the day.
As long as everyone choses a different HH:MM, this should average out pretty nicely.
Erik
Am 01.02.2015 um 20:02 schrieb Sebastian Urbach:
I would like to provide a good service for everyone, even at the end of the month. That's getting harder the more systems are not present at the end of the month.
I could understand the discussion if it were about providing 500 kBit continuously vs. 1 Mbit for 2 of 4 weeks. But the particular case was about providing no less than 6 Mbit continuously, which is easily enough to comfortably browse the web, for doing large downloads and probably exhausts most internet connections in unfree countries. Accordingly it's unlikely a single connection is hobbled by such a bandwidth limitation.
It might be a good idea to relax this recommendation for services above some threshold, where a "limitation" doesn't actually cause a noticably lower quality of service.
Markus
tor-relays@lists.torproject.org
-
Erik de Castro Lopo -
JT Allison -
Markus Hitter -
Sebastian Urbach -
Toralf Förster -
Zack Weinberg