On 3 Sep 2016, at 05:55, grarpamp grarpamp@gmail.com wrote:
On Fri, Sep 2, 2016 at 7:30 AM, Michael Armbruster tor@armbrust.me wrote:
On 2016-09-02 at 13:18, jensm1 wrote:
which shows that the advertised relay bandwidth in the whole network is more than double the actually used bandwidth. While it's certainly nice to have a bit of breathing space to absorb load spikes, I'm wondering,
it's always good to have even more relays or exit nodes, as more "hop points" for connections means more diversity throughout the network
Once a net reaches adequate bandwidth capacity, adding more nodes can do a few things among others... Good:
- Gives operators deployment experience till their bw is needed, at $cost.
- More non-evil relays gives better odds of building a non-evil path, but tor
weight's things so not exactly.
- May add some capacity for directory operations etc
Bad:
- Yields rather unused nodes making it easier for passive
observer to see you tack up and use a path through them, especially if you're crafting paths.
One key here is probably that we don't have a good idea as to the quantity of evil nodes, or the hard interest and real capabilities of PA's.
To make the call you'd need that, and perf metrics of your net under different ratios of advertised:consumed:nodecount, and min/avg/max/stddev of idle/random/full paths, to find any sweet spots / ranges.
Also considerations of impact adding nodes of less bandwidth or more latency than average, versus a campaign to fund replace them.
At 42% util by one metric, it may be money and time better spent elsewhere, even on better qualifying the default 'more nodes good' idea.
There are also latency and contention considerations: an network that runs at 40% capacity has much better latency properties than one at 90% capacity. Also, in the Tor network, as soon as cells are dropped due to any relay being over capacity, this adds a significant delay, because of how Tor retransmits along the entire path, not just the busy relay's hop.
Tim
Tim Wilson-Brown (teor)
teor2345 at gmail dot com PGP C855 6CED 5D90 A0C5 29F6 4D43 450C BA7F 968F 094B ricochet:ekmygaiu4rzgsk6n xmpp: teor at torproject dot org