Estimating the value and cost of the Tor network
Hello everyone, I would like to put together some estimates on bounds of the current value and cost of the capacity of the Tor network as it is, and use that to generate some rough guestimates on what it would cost to grow it (assuming we solved all the complicated problems with respect to maintaining diversity and operator incentives). In order to make these estimates, I need some people to tell me: 1. Your node identity fingerprints. 2. How these fingerprints map to hardware, CPU cores, and uplink. 3. How much you are paying for this uplink per month. If you are paying less than market rate because of a friendly ISP, ideally you would also tell me what the standard market rate is at that ISP. If you have multiple nodes at multiple datacenters, please break these costs and fingerprints out individually rather than telling me aggregate information. I'd most like to hear from people who are doing the recommended best-practice of running one Tor process instance per CPU core, are paying for dedicated uplink at around 100Mbit/sec or higher at a datacenter, and who are not CPU bound on any of their Tor processes. If this does not apply to you, I'd still like to hear from you, but please tell me these details of your setup. Exit node data is most useful, but I am happy to hear about non-exits too. I suspect they may be much cheaper on average, and getting some data on this is also important. Unfortunately, bridges are not that useful at this time. You do not need to send this information publicly to the list. I am happy to receive it privately via GPG. My GPG key id is 0x29846B3C683686CC, and that key signs all of my mail to all torproject lists. You can get it here: https://pgp.mit.edu/pks/lookup?op=get&search=0x29846B3C683686CC -- Mike Perry
Hi, Prices vary widely across different countries. We pay between $400 and $1500 per Gbit/s per month in "popular and cheap locations". In a scenario where we want to grow the network and at least keep the current geographical diversity (or even grow it), we'd have to at least equally strengthen less fortunate locations. The list of the top 20 countries [0] contains countries such as Russia, Poland, Norway and Hungary, where international bandwidth is quite expensive. Hopefully someone from this list can help us compile or find estimates for pricing per country. It would be interesting to learn the prices even for mid-size (~50-100Mbit/s+) relays in any country outside of the top five countries, even more so for crazy places that don't make the top 20. Another factor that we should not ignore is that, while it may be possible to find a set of ISPs that allow mid-size exit relays, it gets harder the more traffic you push, because of the additional workload (and scariness) for the ISP from complaints. [0] https://compass.torproject.org/ -- Moritz Bartl https://www.torservers.net/
Moritz Bartl:
Prices vary widely across different countries. We pay between $400 and $1500 per Gbit/s per month in "popular and cheap locations". In a scenario where we want to grow the network and at least keep the current geographical diversity (or even grow it), we'd have to at least equally strengthen less fortunate locations.
Right. With just one or two identity fingerprints, I can give an estimate on the minimum cost to build an equivalent Tor network with the same capacity, as I have already done. This is not very interesting, though, as you point out. But with just one or two good example identity fingerprints (with pricing) in key locations, I can tell us how much investment it would take to build the Tor network with the diversity we want, using our current load balancing and network load. In other words, I can easily calculate what it would cost to ensure that the network path selection was made up of W% of RU, X% of US relays, Y% of EU relays, Z% of JP relays, etc etc. With many more datapoints I can tell us how much the current Tor network actually costs with its current diversity, but I think that is actually less interesting, unless we wanted to be able to make assumptions like "As soon as we start paying people for bandwidth, all of (or X% of) our volunteers will instantly disappear" (which seems unlikely to me, but others think is a realistic concern). *But* In order to do any of this, I need specific identity fingerprints and prices to do that calculation first. Again, I want to extrapolate from real relays, using our current load balancing. So far only two people have given me identity fingerprints with actual pricing information. I need way more. -- Mike Perry
Hello Perry. 5 TB/month for 20 euro/month at transip in the NL. I run a non-exit relay there. (Thinking about making it an exit.) Kees -- Kees on the move
On 25 Sep 2014, at 03:03, Mike Perry <mikeperry@torproject.org> wrote:
Moritz Bartl:
Prices vary widely across different countries. We pay between $400 and $1500 per Gbit/s per month in "popular and cheap locations". In a scenario where we want to grow the network and at least keep the current geographical diversity (or even grow it), we'd have to at least equally strengthen less fortunate locations.
Right. With just one or two identity fingerprints, I can give an estimate on the minimum cost to build an equivalent Tor network with the same capacity, as I have already done. This is not very interesting, though, as you point out.
But with just one or two good example identity fingerprints (with pricing) in key locations, I can tell us how much investment it would take to build the Tor network with the diversity we want, using our current load balancing and network load.
In other words, I can easily calculate what it would cost to ensure that the network path selection was made up of W% of RU, X% of US relays, Y% of EU relays, Z% of JP relays, etc etc.
With many more datapoints I can tell us how much the current Tor network actually costs with its current diversity, but I think that is actually less interesting, unless we wanted to be able to make assumptions like "As soon as we start paying people for bandwidth, all of (or X% of) our volunteers will instantly disappear" (which seems unlikely to me, but others think is a realistic concern).
*But* In order to do any of this, I need specific identity fingerprints and prices to do that calculation first. Again, I want to extrapolate from real relays, using our current load balancing.
So far only two people have given me identity fingerprints with actual pricing information. I need way more.
-- Mike Perry _______________________________________________ tor-relays mailing list tor-relays@lists.torproject.org https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-relays
I really need identity fingerprints to see how much traffic your node is actually pushing, what its consensus weight is, when and how often it is hibernating, if it is otherwise strangely rate limited, etc. Kees Goossens:
Hello Perry. 5 TB/month for 20 euro/month at transip in the NL. I run a non-exit relay there. (Thinking about making it an exit.) Kees
-- Kees on the move
On 25 Sep 2014, at 03:03, Mike Perry <mikeperry@torproject.org> wrote:
Moritz Bartl:
Prices vary widely across different countries. We pay between $400 and $1500 per Gbit/s per month in "popular and cheap locations". In a scenario where we want to grow the network and at least keep the current geographical diversity (or even grow it), we'd have to at least equally strengthen less fortunate locations.
Right. With just one or two identity fingerprints, I can give an estimate on the minimum cost to build an equivalent Tor network with the same capacity, as I have already done. This is not very interesting, though, as you point out.
But with just one or two good example identity fingerprints (with pricing) in key locations, I can tell us how much investment it would take to build the Tor network with the diversity we want, using our current load balancing and network load.
In other words, I can easily calculate what it would cost to ensure that the network path selection was made up of W% of RU, X% of US relays, Y% of EU relays, Z% of JP relays, etc etc.
With many more datapoints I can tell us how much the current Tor network actually costs with its current diversity, but I think that is actually less interesting, unless we wanted to be able to make assumptions like "As soon as we start paying people for bandwidth, all of (or X% of) our volunteers will instantly disappear" (which seems unlikely to me, but others think is a realistic concern).
*But* In order to do any of this, I need specific identity fingerprints and prices to do that calculation first. Again, I want to extrapolate from real relays, using our current load balancing.
So far only two people have given me identity fingerprints with actual pricing information. I need way more.
-- Mike Perry _______________________________________________ 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
-- Mike Perry
On 25 Sep 2014, at 09:21, Mike Perry <mikeperry@torproject.org> wrote:
I really need identity fingerprints to see how much traffic your node is actually pushing, what its consensus weight is, when and how often it is hibernating, if it is otherwise strangely rate limited, etc.
All nonexits since I started hosting gabelmoo. F8D27B163B9247B232A2EEE68DD8B698695C28DE - 50 Eur/month. 100mbit/s EBE718E1A49EE229071702964F8DB1F318075FF8 - free, 1gbps F2044413DAC2E02E3D6BCF4735A19BCA1DE97281 - free, 1gbps None of my nodes come close to saturate either CPU nor their connection. Cheers Sebastian
Sorry here it is: 5AA81AE1F2584F1740D813F407FF08028783C5DF Note that it's been operational (and ramping up) for two months only. But there are some other relays at the same VPS that look like they may have the same pricing package. (Source: http://torstatus.blutmagie.de, search on transip.net) DD77 4E2C 2D39 3C0E EDC5 8C98 4181 E153 283F 8E17 7FF7 CAFE B36B FEA8 5942 D7C5 5E3E 81F2 6A15 DE9E Kees On 25 Sep 2014, at 09:21, Mike Perry <mikeperry@torproject.org> wrote:
I really need identity fingerprints to see how much traffic your node is actually pushing, what its consensus weight is, when and how often it is hibernating, if it is otherwise strangely rate limited, etc.
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Hi Mike, What follows is the data from most of our nodes. We've had a few problems with some so one or two aren't included (nor are the seized ones since their fingerprints have now been blacklisted). Hope it helps. Chandler01 & Chandler02 37.148.163.38 Chandler01 D78AB0013D95AFA60757333645BAA03A169DF722 Chandler02 6F545A39D4849C9FE5B08A6D68C8B3478E4B608B CPU Usage: 80-90% (per process) CPU: 1x Dual Core i3 550 Cost: 456.61 Euro/month Chandler03 & Chandler04 128.204.203.103 Chandler03 5E87B10B430BA4D9ADF1E1F01E69D3A137FB63C9 Chandler04 0824CE7D452B892D12E081D36E7415F85EA9988F CPU Usage: 80% (per process) CPU: 1x Dual Core i3 550 Cost: 456.61 Euro/month Chandler05 & Chandler06 5.104.224.246 Chandler05 35961469646A623F9EE03B7B45296527A624AAFD Chandler06 1EA968C956FBC00617655A35DA872D319E87C597 CPU Usage: 90% (05) & 70% (06) CPU: 1x Dual Core i3 550 Cost: 430.04 Euro/month Chandler07 & Chandler08 89.207.132.76 Chandler07 E5A21C42B0FDB88E1A744D9A0388EFB2A7A598CF Chandler08 5D1CB4B3025F4D2810CF12AB7A8DDDD6FC10F139 CPU Usage: 90% (07) & 60% (08) CPU: 1x Dual Core i3 550 Cost: 456.61 Euro/month Chandler09 & Chandler10 77.95.231.11 Chandler09 722B4DF4848EC8C15302C7CF75B52C65BAE3843A Chandler10 93CD9231C260558D77331162A5DC5A4C692F5344 CPU Usage: 70-80% (per process) CPU: 1x Dual Core i3 550 Cost: 350.36 Euro/month Chandler21 77.95.224.187 Chandler21 A3C3D2664F5E92171359F71931AA2C0C74E2E65C CPU Usage: 80% CPU: 1x Dual Core i3-2100 Cost: 244.03 Euro/month Chandler22 89.207.128.241 Chandler22 575B40EF095A0F2B13C83F8485AFC56453817ABF CPU Usage: 70% CPU: 1x Dual Core i3-2100 Cost: 244.03 Euro/month Chandler23 5.104.224.15 Chandler23 1A334E26E66303A4DE046E8733942601F5523E62 CPU Usage: 55% CPU: 1x Dual Core i3-2100 Cost: 244.04 Euro/month Chandler25 1324EC51FBFA5FD1A11B94563E8D2A7999CD8F57 CPU: Intel(R) Core(TM) i7 CPU 950 @ 3.07GHz quad core Network speed: 1gbps port CPU usage: 92% to 96% (one instance of Tor running) Cost: 373.33 Euro/month ChandlerRelay01 7E45A9F483D87A37C9412575DCB723EC1FF02AB0 Location: Latvia Type: Virtual private server / Virtualization: OpenVZ Operating system: CentOS 7 CPU: Intel(R) Xeon(R) CPU E5-1620 v2 @ 3.70GHz quad core on the host virtualization is OpenVZ so we don't know how much is dedicated to us Network speed: 100mbps port, but bandwidth throttled by provider to 5mbps CPU usage: 5% to 12% Cost: 7 euros per month, but we took it for 1 year and paid 70 euros per year. ChandlerRelay02 A7C7EAFE7813F87B17A2D000913E73A46BB41B33 Location: Latvia Type: Virtual private server / VIrtualization: OpenVZ Operating system: CentOS 7 CPU: Intel(R) Xeon(R) CPU E5-1620 v2 @ 3.70GHz quad core on the host virtualization is OpenVZ so we don't know how much is dedicated to us Network speed: 100mbps port, but bandwidth throttled by provider to 5mbps CPU usage: 5% to 12% Cost: 7 euros per month, but we took it for 1 year and paid 70 euros per year. ChandlerRelay04 1A27FACDE82EEBC2018A7795277932C2ECBF105A Location: Latvia Type: Virtual private server / VIrtualization: OpenVZ Operating system: Debian Wheezy CPU: Intel(R) Xeon(R) CPU E5-1620 v2 @ 3.70GHz quad core on the host virtualization is OpenVZ so we don't know how much is dedicated to us Network speed: 100mbps port, but bandwidth throttled by provider to 5mbps CPU usage: 5% to 18% Cost: 7 euros per month, but we took it for 1 year and paid 70 euros per year. ChandlerRelay05 4A596EC7459E25791F07A84CB977D2CCFBB4247C Location: Latvia Type: Virtual private server / VIrtualization: OpenVZ Operating system: Debian Wheezy CPU: Intel(R) Xeon(R) CPU E5-1620 v2 @ 3.70GHz quad core on the host virtualization is OpenVZ so we don't know how much is dedicated to us Network speed: 100mbps port, but bandwidth throttled by provider to 5mbps CPU usage: 5% to 20% Cost: 7 euros per month, but we took it for 1 year and paid 70 euros per year. marmellata FE89FDC5DC618FA677C4B85F702D2CFCFFC48170 Location: Sweden Type: Virtual private server / VIrtualization: OpenVZ Operating system: Debian Wheezy CPU: Intel(R) Xeon(R) CPU E5-1620 v2 @ 3.70GHz quad core on the host virtualization is OpenVZ so we don't know how much is dedicated to us Network speed: 100mbps port, but bandiwdth throttled by provider to 5mbps CPU usage: 10% to 30% Cost: 7 euros per month, but we took it for 1 year and paid 70 euros per year. We additionally run 21 obfuscated 100mbit obfuscated bridges (do you want info on these too?) Regards, Thomas White On 25/09/2014 08:21, Mike Perry wrote:
I really need identity fingerprints to see how much traffic your node is actually pushing, what its consensus weight is, when and how often it is hibernating, if it is otherwise strangely rate limited, etc.
Kees Goossens:
Hello Perry. 5 TB/month for 20 euro/month at transip in the NL. I run a non-exit relay there. (Thinking about making it an exit.) Kees
-- Kees on the move
On 25 Sep 2014, at 03:03, Mike Perry <mikeperry@torproject.org> wrote:
Moritz Bartl:
Prices vary widely across different countries. We pay between $400 and $1500 per Gbit/s per month in "popular and cheap locations". In a scenario where we want to grow the network and at least keep the current geographical diversity (or even grow it), we'd have to at least equally strengthen less fortunate locations.
Right. With just one or two identity fingerprints, I can give an estimate on the minimum cost to build an equivalent Tor network with the same capacity, as I have already done. This is not very interesting, though, as you point out.
But with just one or two good example identity fingerprints (with pricing) in key locations, I can tell us how much investment it would take to build the Tor network with the diversity we want, using our current load balancing and network load.
In other words, I can easily calculate what it would cost to ensure that the network path selection was made up of W% of RU, X% of US relays, Y% of EU relays, Z% of JP relays, etc etc.
With many more datapoints I can tell us how much the current Tor network actually costs with its current diversity, but I think that is actually less interesting, unless we wanted to be able to make assumptions like "As soon as we start paying people for bandwidth, all of (or X% of) our volunteers will instantly disappear" (which seems unlikely to me, but others think is a realistic concern).
*But* In order to do any of this, I need specific identity fingerprints and prices to do that calculation first. Again, I want to extrapolate from real relays, using our current load balancing.
So far only two people have given me identity fingerprints with actual pricing information. I need way more.
-- Mike Perry _______________________________________________ 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
_______________________________________________ tor-relays mailing list tor-relays@lists.torproject.org https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-relays
-----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.22 (MingW32) iQIcBAEBAgAGBQJUJJVJAAoJEE2uQiaesOsL6dYP/RtqY5TvFcefbfTeXnlsf8mQ cL3HrLVINqmXqyUTX3/dx3VdskSEzOUpncRDVqUh1SdTqQZzGz6Ty+m467hQWSPi rN1j/ofqpwMGqpwhWIKAOtFSj2uAsfw9u2mXigBo+E1kI8oIh7GG/M6hmAZvkvJv JVjf7yQSZsUKuazR+hHQfGR24s26mWH8iM3orAzKh/WAE6JZ10aC69V/itj1BILq uVdbGtPt9Ym61QeIwvrGOCYMvEB+kFTtF81j7Vbn3ui/FOKAoS0FzOpyNCn1o/Ea eEwZRDXal9bpJe3YwyIVR5NxpoOHbH2JVzBmZ2BsTw7nh5fKmd2RMzXRVDEXqOoY JkRl0w73YAhszBKoA9n6e4E4cLxhzb8NfuBrjkPhIq0Npi9Ge09guzzDL2JVcezL YAzNr+vUzHGHCJcxzwL+UBSoX87z6iV4xhOdOL3XO1mthvbipRIo8BXMNwMB3Pqm glywpkh2XsSOKdO03MDHQxaG+7r76D06kAc6sucV/6ol4MHjHk0/ebXHs+DvDKPL XQk/mIP9kOegpFgbS3k0nROK1XXZrrjFEQYicS+nU4sU0k5d0eVKlvG0uy/jmk9W ooBBC+R/thhJ5y90klgVOH4TASFbR5HCHyhIqi6+4edSCD/277rCJu2m9nZY7U2w ZrwCHty5Radt851Is94a =BHxG -----END PGP SIGNATURE-----
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Being paid for doing what I love ??? That's really a great deal !!! And I believe most people is thinking the same. Lluis Spain On 09/25/2014 03:03 AM, Mike Perry wrote:
Moritz Bartl:
Prices vary widely across different countries. We pay between $400 and $1500 per Gbit/s per month in "popular and cheap locations". In a scenario where we want to grow the network and at least keep the current geographical diversity (or even grow it), we'd have to at least equally strengthen less fortunate locations.
Right. With just one or two identity fingerprints, I can give an estimate on the minimum cost to build an equivalent Tor network with the same capacity, as I have already done. This is not very interesting, though, as you point out.
But with just one or two good example identity fingerprints (with pricing) in key locations, I can tell us how much investment it would take to build the Tor network with the diversity we want, using our current load balancing and network load.
In other words, I can easily calculate what it would cost to ensure that the network path selection was made up of W% of RU, X% of US relays, Y% of EU relays, Z% of JP relays, etc etc.
With many more datapoints I can tell us how much the current Tor network actually costs with its current diversity, but I think that is actually less interesting, unless we wanted to be able to make assumptions like "As soon as we start paying people for bandwidth, all of (or X% of) our volunteers will instantly disappear" (which seems unlikely to me, but others think is a realistic concern).
*But* In order to do any of this, I need specific identity fingerprints and prices to do that calculation first. Again, I want to extrapolate from real relays, using our current load balancing.
So far only two people have given me identity fingerprints with actual pricing information. I need way more.
_______________________________________________ tor-relays mailing list tor-relays@lists.torproject.org https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-relays
-----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.12 (GNU/Linux) iQEcBAEBAgAGBQJUI+epAAoJENx83SfoKigqq8QH+wW54+yuCCfR5t4Voec93miN 775t4Gr/PBkHwwqPCE68quooT94wGlFn+ADsd8w8PL9X5eL6NUzHe+DFaf7vufFI rhOUqZqXob2GHS8hT8kZ/4N3x+n9ZhDFzsckA+aM0IopLRgzHUa9t73uxFIDtoKw hKu+7N+4Ua13nGQSgOmf8092vQpDRLgyAANhY78zAHtWLS+gQ9X/QENsuR6ZrFij zrgDn35F8WVzMbFeeWC1nxk2iKgVhyqH5A3/5ilhsgaWsInu3Ey5guv2yTuqgjxa freDu4EFNmDTp3ikxmYkLEWwFhfMllykQ8nLodd9kRCxgauZFMpUQreclcmr6gs= =omge -----END PGP SIGNATURE-----
Am 25.09.2014 um 03:03 schrieb Mike Perry:
So far only two people have given me identity fingerprints with actual pricing information. I need way more.
62CD4F495DB0A5C5C74B3D39722BFA1AFEF365C9 28,00 Euro/Month 10 TB Traffic with 100 MBit, after that limited to 10 MBit -- Mit freundlichen Grüßen Tobias Bauer
Hello Mike, I am running exit relay (was non-exit in the past) - F12AFDB3FEC184E76944579579F762F1142C7ED2 Its one core VPS with 1GB of RAM on 100 Mbps line. I pay 1016 CZK, which is about 47 USD, per year. Best, dope457 Dne 2014-09-24 01:11, Mike Perry napsal:
Hello everyone,
I would like to put together some estimates on bounds of the current value and cost of the capacity of the Tor network as it is, and use that to generate some rough guestimates on what it would cost to grow it (assuming we solved all the complicated problems with respect to maintaining diversity and operator incentives).
In order to make these estimates, I need some people to tell me: 1. Your node identity fingerprints. 2. How these fingerprints map to hardware, CPU cores, and uplink. 3. How much you are paying for this uplink per month.
If you are paying less than market rate because of a friendly ISP, ideally you would also tell me what the standard market rate is at that ISP.
If you have multiple nodes at multiple datacenters, please break these costs and fingerprints out individually rather than telling me aggregate information.
I'd most like to hear from people who are doing the recommended best-practice of running one Tor process instance per CPU core, are paying for dedicated uplink at around 100Mbit/sec or higher at a datacenter, and who are not CPU bound on any of their Tor processes. If this does not apply to you, I'd still like to hear from you, but please tell me these details of your setup.
Exit node data is most useful, but I am happy to hear about non-exits too. I suspect they may be much cheaper on average, and getting some data on this is also important. Unfortunately, bridges are not that useful at this time.
You do not need to send this information publicly to the list. I am happy to receive it privately via GPG. My GPG key id is 0x29846B3C683686CC, and that key signs all of my mail to all torproject lists. You can get it here: https://pgp.mit.edu/pks/lookup?op=get&search=0x29846B3C683686CC
_______________________________________________ tor-relays mailing list tor-relays@lists.torproject.org https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-relays
participants (8)
-
dope457 -
Kees Goossens -
Lluís -
Mike Perry -
Moritz Bartl -
Sebastian Hahn -
Thomas White -
Tobias Bauer [TB-ITF]