Which OS gives usually the best performance for a relay?
Hello, I am thinking of setting up a new relay. I know that the hardware in the server is going to be the bottleneck, not my Internet connection. I have a problem deciding on which OS to use for the relay. A few years ago when I had a similar relay going, I had it running on OpenBSD first. Then I changed the OS to FreeBSD and the performance got about 20% better. I have no idea if this would be the case today too. So I think that maybe it's either FreeBSD or Debian that would be "best", but I have nothing concrete to base that decision on unless I try them both. I am going to use a Via C7 board in this specific case. So I suspect that it's the maturity of the VIA drivers in the OS that is going to make the difference. Still I would like to know how to think in similar situations in the future even for other hardware. Has anyone any concrete experience of the tor relay speeds on different operating systems? I don't want to start a flame war of religious beliefs, but I suspect that OSes differ in how optimized they are for different tasks. Thankful for any constructive input on this. Regards, Farid
Hi! All modern Operating Systems should be up to the task of running a Tor relay, if configured right. The question about which one will work best has probably no general answer, but will depend on the hardware (and software) configuration used, the quality of the drivers for your specific hardware, etc. and you'll have to try for yourself which one will give you the best performance. But please keep in mind that diversity is also very important, since an overwhelming majority of relays runs on Linux. So even if(!) Debian would be a bit faster, it could very well be worth it to sacrifice a few percent performance to increase the OS diversity. Am 06.09.2016 um 22:14 schrieb Farid Joubbi:
Hello,
I am thinking of setting up a new relay.
I know that the hardware in the server is going to be the bottleneck, not my Internet connection.
I have a problem deciding on which OS to use for the relay.
A few years ago when I had a similar relay going, I had it running on OpenBSD first.
Then I changed the OS to FreeBSD and the performance got about 20% better.
I have no idea if this would be the case today too.
So I think that maybe it's either FreeBSD or Debian that would be "best", but I have nothing concrete to base that decision on unless I try them both.
I am going to use a Via C7 board in this specific case. So I suspect that it's the maturity of the VIA drivers in the OS that is going to make the difference. Still I would like to know how to think in similar situations in the future even for other hardware.
Has anyone any concrete experience of the tor relay speeds on different operating systems?
I don't want to start a flame war of religious beliefs, but I suspect that OSes differ in how optimized they are for different tasks.
Thankful for any constructive input on this.
Regards,
Farid
_______________________________________________ tor-relays mailing list tor-relays@lists.torproject.org https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-relays
I had not thought of the diversity that way. Thanks for pointing it out. I am still interested in the subject though, if anyone has any specific examples of some kind of general rules of why one OS usually performs better than some other OS as a tor relay... I realize that I might not get any good answers since my question is kind of broad and unspecific. ________________________________ From: tor-relays <tor-relays-bounces@lists.torproject.org> on behalf of jensm1 <jensm1@bbjh.de> Sent: 07 September 2016 13:08 To: tor-relays@lists.torproject.org Subject: Re: [tor-relays] Which OS gives usually the best performance for a relay? Hi! All modern Operating Systems should be up to the task of running a Tor relay, if configured right. The question about which one will work best has probably no general answer, but will depend on the hardware (and software) configuration used, the quality of the drivers for your specific hardware, etc. and you'll have to try for yourself which one will give you the best performance. But please keep in mind that diversity is also very important, since an overwhelming majority of relays runs on Linux. So even if(!) Debian would be a bit faster, it could very well be worth it to sacrifice a few percent performance to increase the OS diversity. Am 06.09.2016 um 22:14 schrieb Farid Joubbi: Hello, I am thinking of setting up a new relay. I know that the hardware in the server is going to be the bottleneck, not my Internet connection. I have a problem deciding on which OS to use for the relay. A few years ago when I had a similar relay going, I had it running on OpenBSD first. Then I changed the OS to FreeBSD and the performance got about 20% better. I have no idea if this would be the case today too. So I think that maybe it's either FreeBSD or Debian that would be "best", but I have nothing concrete to base that decision on unless I try them both. I am going to use a Via C7 board in this specific case. So I suspect that it's the maturity of the VIA drivers in the OS that is going to make the difference. Still I would like to know how to think in similar situations in the future even for other hardware. Has anyone any concrete experience of the tor relay speeds on different operating systems? I don't want to start a flame war of religious beliefs, but I suspect that OSes differ in how optimized they are for different tasks. Thankful for any constructive input on this. Regards, Farid _______________________________________________ tor-relays mailing list tor-relays@lists.torproject.org<mailto:tor-relays@lists.torproject.org> https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-relays
On 09/07/16 11:54, Farid Joubbi wrote:
I had not thought of the diversity that way.
There's a host of diversity issues with Tor to cover, but I tend to think OS diversity is one of the more critical. These are some reports we generate at TDP: https://torbsd.github.io/dirty-stats.html
Thanks for pointing it out.
I am still interested in the subject though, if anyone has any specific examples of some kind of general rules of why one OS usually performs better than some other OS as a tor relay...
There's lots of factors to consider once one is using similar hardware, the same bandwidth and pipes, etc. The ultimate difficulty in doing a test comparison is creating identical scenarios. Tor is a more or less random anonymity routing network, which breaks any notion of repeatability. This piece makes a general case pretty clearly: http://queue.acm.org/detail.cfm?id=2732268 However, the FreeBSD networking stack is known to be extremely fast and optimized. OpenBSD is built to be secure by default, so many default knobs are aimed at keeping a live system secure. The most obvious parameter to adjust is kern.maxfiles in the sysctls. Certainly both are underrepresented in the Tor public network... .but we're working on it. . .
I realize that I might not get any good answers since my question is kind of broad and unspecific.
Clearly you are asking the right questions, which is what's critical IMHO. Not directing to the OP, but I also strongly think one should stick with the OS they are most comfortable in administering, regardless of diversity questions. If someone's never used a Unix-like system before, and can't manage to edit a file with vi(1), start elsewhere :) g -- 5F77 765E 40D6 5340 A0F5 3401 4997 FF11 A86F 44E2
George:
On 09/07/16 11:54, Farid Joubbi wrote:
I had not thought of the diversity that way.
There's a host of diversity issues with Tor to cover, but I tend to think OS diversity is one of the more critical.
These are some reports we generate at TDP:
some more stats using onionoo data and mainly based on consensus weight fraction (IMHO more relevant than relay count - Windows is a good example, have a look at relay count vs. consensus weight fraction) https://github.com/ornetstats https://github.com/ornetstats/stats/blob/master/o/os_share.txt
On 09/07/2016 10:34 AM, George wrote:
On 09/07/16 11:54, Farid Joubbi wrote:
I had not thought of the diversity that way.
There's a host of diversity issues with Tor to cover, but I tend to think OS diversity is one of the more critical.
With apologies to Akira Kurosawa, I think of this as the Seven Samurai Problem. Diverse samurai are a given; their optimum deployment is not.
The ultimate difficulty in doing a test comparison is creating identical scenarios. Tor is a more or less random anonymity routing network, which breaks any notion of repeatability.
The Tor network is uniquely protean.
I realize that I might not get any good answers since my question is kind of broad and unspecific.
Clearly you are asking the right questions, which is what's critical IMHO.
Not directing to the OP, but I also strongly think one should stick with the OS they are most comfortable in administering, regardless of diversity questions. If someone's never used a Unix-like system before, and can't manage to edit a file with vi(1), start elsewhere :)
I couldn't agree more. I don't get under the hood much; I rely on Ubuntu Linux and Tor's default configuration. Other relays are much more finely tuned, I'm sure, on machinery replete with all the bells and whistles. In this regard the old engineering dictum that "The best is the enemy of the good enough" (which itself dates to Voltaire) is pertinent. We have a diverse ecosystem of volunteers, which perforce deploys a host of heterogeneous attack surfaces. I'm not going to tweak what I don't understand. I'm not a Kyūzō; I'm a Heihachi.
participants (5)
-
Farid Joubbi -
George -
jensm1 -
Kenneth Freeman -
nusenu