Re: [tor-relays] Question regarding exit sizing

[igor.n.mitrofanov]

This is why I chose to run multiple 100 Mhz nodes. I over-advertise the 
capacity, so that some of them get a bit more traffic than they really should, 
because the bwauths under-measure NA relays and, worse, keep changing their 
measurements for no apparent reason. On the same bare metal box one relay can 
be measured as 3-4x another, easily.
 Original message From: Conrad Rockenhaus 
<con...@rockenhaus.com> Date: 2/10/18  12:33 AM  (GMT-08:00) To: 
tor-relays@lists.torproject.org Subject: Re: [tor-relays] Question regarding 
exit sizing 
The 500 Mbps instance would either be one of my private servers in my co-lo, or 
a dedicated server in one of my private cloud hosting locations. With both 
contacts, bandwidth costs aren’t an issue, but if one big instance would work I 
would put it on the same hardware that I am running a server that averages 
about 700+ mbps consistently.  The small servers that I’m hitting 100 mbps on, 
I’m just getting low cost VPSes for since….they do the job and they do the job 
well, they’re guaranteed a minimum of 100mbps bandwidth to the first tier 1 hop 
and Atlas shows them consistently used at that level, so I’m happy.
I mainly wanted to give back big to the community because Tor gave me the idea 
for my latest free for personal/charge for business use idea that I’m going to 
roll out soon (I’ll gladly send y’all a link, as I think it’s something that 
would be very useful).

I haven’t noticed any bad measurements…the three relays I run now, well, one 
just started this week so we can throw that one out for now, but the other two  
are showing 12.55 MiB/s and 12.28 MiB/s, and I’m guaranteed 100 mbps, so I’m 
doing pretty well on those two. Since the priority is exit nodes, I’ll probably 
add two more exit nodes in Canada, leaving four exits, and one relay there.
But I do get your points, and the more I do think about it, it would be better 
to just spread it all out, so I guess whenever I start spinning up nodes in 
Europe I’ll just use VPSes. One other thing I forgot to realize is I’m seeing a 
steady increase in the amount of DDoS attacks on my exits as of late. My 
provider tries to mitigate them as much as possible, but it’s annoying for the 
end users going through the node and it’s annoying for the people who are 
getting affected by the DDoS. Putting everything on one big box is just 
screaming “Here, attack me right here plz, kthx."


On Feb 10, 2018, at 1:44 AM, tor <t...@anondroid.com> wrote:
What scenario is better for the network - adding five 100mbps nodes, or one 500 
mbps node?


Are we talking bare metal or VPS? A VPS will probably bottleneck on RAM or CPU 
before hitting 500 Mpbs.

Bare metal would stand a chance with the right hardware and tuning, but I 
wouldn't assume you'll hit 500 Mbps on any given node.

Due to the nature of the bandwidth measurements, physical location matters too. 
You're at the mercy of Tor's bandwidth authorities and in my experience, the 
further away from Europe, the worse your measurements will be, and so again you 
may not hit 500 Mbps.

Basically, you shouldn't assume that whatever bandwidth you plan for and 
advertise will come your way. 

I think you'd have better luck with 5x 100 Mbps nodes, or maybe 3x 200 Mbps 
nodes. You can also run 2 relays per IP.

There are advantages to spreading out the load (like redundancy). I also think 
Tor's bandwidth measurements and consensus weights are fickle, and some of the 
variables are out of your control (what else is going on in your rack, 
datacenter, upstream, etc.). You could use ansible-relayor to turn up a bunch 
of nodes, wait to see which ones are the most performant, and then keep the 
best ones. That's what I would do. :)


___
tor-relays mailing list
tor-relays@lists.torproject.org
https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-relays

___
tor-relays mailing list
tor-relays@lists.torproject.org
https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-relays

Re: [tor-relays] Question regarding exit sizing

[Conrad Rockenhaus]

The 500 Mbps instance would either be one of my private servers in my co-lo, or 
a dedicated server in one of my private cloud hosting locations. With both 
contacts, bandwidth costs aren’t an issue, but if one big instance would work I 
would put it on the same hardware that I am running a server that averages 
about 700+ mbps consistently.  The small servers that I’m hitting 100 mbps on, 
I’m just getting low cost VPSes for since….they do the job and they do the job 
well, they’re guaranteed a minimum of 100mbps bandwidth to the first tier 1 hop 
and Atlas shows them consistently used at that level, so I’m happy.

I mainly wanted to give back big to the community because Tor gave me the idea 
for my latest free for personal/charge for business use idea that I’m going to 
roll out soon (I’ll gladly send y’all a link, as I think it’s something that 
would be very useful).

I haven’t noticed any bad measurements…the three relays I run now, well, one 
just started this week so we can throw that one out for now, but the other two  
are showing 12.55 MiB/s and 12.28 MiB/s, and I’m guaranteed 100 mbps, so I’m 
doing pretty well on those two. Since the priority is exit nodes, I’ll probably 
add two more exit nodes in Canada, leaving four exits, and one relay there.

But I do get your points, and the more I do think about it, it would be better 
to just spread it all out, so I guess whenever I start spinning up nodes in 
Europe I’ll just use VPSes. One other thing I forgot to realize is I’m seeing a 
steady increase in the amount of DDoS attacks on my exits as of late. My 
provider tries to mitigate them as much as possible, but it’s annoying for the 
end users going through the node and it’s annoying for the people who are 
getting affected by the DDoS. Putting everything on one big box is just 
screaming “Here, attack me right here plz, kthx."




> On Feb 10, 2018, at 1:44 AM, tor  wrote:
> 
>> What scenario is better for the network - adding five 100mbps nodes, or one 
>> 500 mbps node?
> 
> 
> Are we talking bare metal or VPS? A VPS will probably bottleneck on RAM or 
> CPU before hitting 500 Mpbs.
> 
> Bare metal would stand a chance with the right hardware and tuning, but I 
> wouldn't assume you'll hit 500 Mbps on any given node.
> 
> Due to the nature of the bandwidth measurements, physical location matters 
> too. You're at the mercy of Tor's bandwidth authorities and in my experience, 
> the further away from Europe, the worse your measurements will be, and so 
> again you may not hit 500 Mbps.
> 
> Basically, you shouldn't assume that whatever bandwidth you plan for and 
> advertise will come your way. 
> 
> I think you'd have better luck with 5x 100 Mbps nodes, or maybe 3x 200 Mbps 
> nodes. You can also run 2 relays per IP.
> 
> There are advantages to spreading out the load (like redundancy). I also 
> think Tor's bandwidth measurements and consensus weights are fickle, and some 
> of the variables are out of your control (what else is going on in your rack, 
> datacenter, upstream, etc.). You could use ansible-relayor to turn up a bunch 
> of nodes, wait to see which ones are the most performant, and then keep the 
> best ones. That's what I would do. :)
> 
> 
> ___
> tor-relays mailing list
> tor-relays@lists.torproject.org
> https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-relays

___
tor-relays mailing list
tor-relays@lists.torproject.org
https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-relays

Re: [tor-relays] Question regarding exit sizing

[tor]

> What scenario is better for the network - adding five 100mbps nodes, or one 
> 500 mbps node?


Are we talking bare metal or VPS? A VPS will probably bottleneck on RAM or CPU 
before hitting 500 Mpbs.

Bare metal would stand a chance with the right hardware and tuning, but I 
wouldn't assume you'll hit 500 Mbps on any given node.

Due to the nature of the bandwidth measurements, physical location matters too. 
You're at the mercy of Tor's bandwidth authorities and in my experience, the 
further away from Europe, the worse your measurements will be, and so again you 
may not hit 500 Mbps.

Basically, you shouldn't assume that whatever bandwidth you plan for and 
advertise will come your way. 

I think you'd have better luck with 5x 100 Mbps nodes, or maybe 3x 200 Mbps 
nodes. You can also run 2 relays per IP.

There are advantages to spreading out the load (like redundancy). I also think 
Tor's bandwidth measurements and consensus weights are fickle, and some of the 
variables are out of your control (what else is going on in your rack, 
datacenter, upstream, etc.). You could use ansible-relayor to turn up a bunch 
of nodes, wait to see which ones are the most performant, and then keep the 
best ones. That's what I would do. :)


___
tor-relays mailing list
tor-relays@lists.torproject.org
https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-relays