Wow, that AES-NI is hot shit! Well, let me know how the US folks here can help. I'd offer to be the host for the kickstarter project for this, but I need to check into some legal stuff on my end before I commit to that. The configuration stuff below sounds good to me, and it sounds like you've thought it out. At what point can torservers be confident it can support $800/month? (I don't intend for you to answer that, just that this should be more of a non-profit business calculation with a certain assumed level of steady donators so that no payments are missed, etc. ... which makes the yearly price very attractive, assuming a refund if they get complaints that can only be solved by killing the node.) best, Joe On Tuesday, February 8, 2011, Moritz Bartl <moritz@xxxxxxxxxxxxxx> wrote: > Hi, > >> Nice. One strategy might be to set this up on a month-to-month until >> there's enough money for buying a full year. > > I'm trying to get that setup fee waived, too. I'm not sure if we ever > want to pay one year up front, especially because they try to avoid the > possible implications of Tor on their network. But I think setting this > up is worth a try - depending on the setup fee. > >> A kickstarter project > > Yes, I have been planning this from the start. Unfortunately, you cannot > create Kickstarter projects from outside US yet. :-/ > > http://kickstarter.uservoice.com/forums/10955-general/suggestions/186005-expand-kickstarter-to-international-project-creato > > I also want to start a Kickstarter project for the first Icelandic exit. > I've been in contact with the owner of ThorDC (www.thordc.is), who wants > 12€/month per Mbps. > >> Any idea how you'd configure this Mo? One >> massive node or 4-5 200 Mbps nodes? Or tons of tiny nodes all >> Family'd together? > > I have pretty good experience with our Softlayer node, which was on a > Gbit line too. Mike Perry is reportedly doing ~800Mbit/s on their > server. I have specifically asked for a CPU with AES-NI support, which > is why I am particularly interested in this offer: the Xeon E5620 > supports AES-NI, which should lower CPU load considerably (and I would > love to play with it and let the Tor community know of the results). > > http://en.wikipedia.org/wiki/AES-NI > > We experimented with all popular virtualization solutions, but ended up > putting the Tor processes directly on the host, because neither Xen, > OpenVZ nor KVM were able to handle the amount of connections coming in > without impacting performance heavily. Because Tor still doesn't scale > on multicore CPUs yet, I would run one logical Tor node for each CPU > core. All in the same family of course - all our nodes are in one > family. I want to keep the number as low as possible though, because > high speed nodes are more likely to be noticed on TorStatus/by the Tor > community than a hundred smaller ones. :) And it probably doesn't help > to run multiple Tor processes per core. > > -- > Moritz Bartl > http://www.torservers.net/ > > -- Joseph Lorenzo Hall ACCURATE Postdoctoral Research Associate UC Berkeley School of Information Princeton Center for Information Technology Policy http://josephhall.org/