On Mon, 18 Feb 2019 09:06:05 -0500 Neel Chauhan neel@neelc.org wrote:
Verizon gives both 300 mbps upload and download speeds. Uploads are more heavily oversubscribed on FiOS, primarily because GPON gives 2.5gbps downloads and 1.25gbps uploads.
But then again the upload will be barely utilized by typical residential Internet users.
Still my recommendation is to test your bandwidth in multiple ways first, be it speedtest.net, or (better yet) https://github.com/sivel/speedtest-cli, or iperf3 servers, if you can find any near your location.
If tests show that you do get near 300 Mbit both directions, the next step would be to just set up two instances of Tor, as I suggested before in your thread[1]. Actually fun to see my prediction from back then coming true precisely (with regard to getting only 200 Mbit).
[1] https://www.mail-archive.com/tor-relays@lists.torproject.org/msg15819.html
Running two instances is the universal solution which should improve Tor's bandwidth utilization on almost any connection.
Roman,
But then again the upload will be barely utilized by typical residential Internet users.
True.
Still my recommendation is to test your bandwidth in multiple ways first, be it speedtest.net, or (better yet) https://github.com/sivel/speedtest-cli, or iperf3 servers, if you can find any near your location.
I am getting 300 Mbps in both directions.
If tests show that you do get near 300 Mbit both directions, the next step would be to just set up two instances of Tor, as I suggested before in your thread[1]. Actually fun to see my prediction from back then coming true precisely (with regard to getting only 200 Mbit).
[1] https://www.mail-archive.com/tor-relays@lists.torproject.org/msg15819.html
Being capped at 200 Mbps was because `powerd` wasn't enabled on my FreeBSD, and "turbo" frequencies weren't being used. Enabling `powerd` means I feel my relay can handle 300 Mbps (and CPU usage dropped because the clock speed increased). Previously 10 MB/s (80 Mbps) took 30% of CPU, now the same amount of bandwidth takes 20%.
Running two instances is the universal solution which should improve Tor's bandwidth utilization on almost any connection.
I'll look at this.
I feel it's my Linksys WRT1900AC because consumer routers aren't designed for the traffic high-bandwidth Tor relays handle, even after flashing things like OpenWrt.
Also see: https://arstechnica.com/gadgets/2016/09/the-router-rumble-ars-diy-build-face...
Would running two instances help with a consumer router's limited NAT Table?
-Neel
===
On Mon, 18 Feb 2019 13:05:47 -0500 Neel Chauhan neel@neelc.org wrote:
Being capped at 200 Mbps was because `powerd` wasn't enabled on my FreeBSD, and "turbo" frequencies weren't being used. Enabling `powerd` means I feel my relay can handle 300 Mbps (and CPU usage dropped because the clock speed increased). Previously 10 MB/s (80 Mbps) took 30% of CPU, now the same amount of bandwidth takes 20%.
If you are not bumping into 100%, it shouldn't matter, especially at levels like 30% or 20%, with plenty of headroom left.
I feel it's my Linksys WRT1900AC because consumer routers aren't designed for the traffic high-bandwidth Tor relays handle, even after flashing things like OpenWrt.
Also see: https://arstechnica.com/gadgets/2016/09/the-router-rumble-ars-diy-build-face...
Would running two instances help with a consumer router's limited NAT Table?
It is true that running two instances will multiply by 2x the amount of connections the router has to keep track of.
There is no "limited NAT table" per se with OpenWRT, just whatever fits into RAM (and will be progressively slower as CPU time to process it increases with table size).
Take a look at what kind of loads the router currently experiences, things such as average CPU load and RAM usage. OpenWRT shows both nicely even in the web UI. The WRT1900AC has plenty of RAM (256 or 512MB depending on revision) and a dual core CPU, so there is some hope that it is not currently the bottleneck, and won't be even with two instances of Tor.
tor-relays@lists.torproject.org