On Mon 21 Dec 2015 09:00:43 PM ED Fochler wrote: > 120/5 mbps connection is VERY asymmetric, and implies cable. Cable > companies often prioritize traffic in unusual ways. Even if they don’t, > the acks going upstream for payload coming down could saturate your > upstream without uploading any content at all. I would put limiters on > your traffic to make sure that you stay below saturation upstream and down, > allowing any local packet queueing or prioritization to actually matter. > If you saturate your connection, then you’re letting the upstream provider > decide what traffic to throw away. > > I have had no problem with bitTorrent behind a PFSense soekris 6501 or on > OpenBSD on the soekris on 25/15 or 50/50 connections. When using OpenBSD I > had traffic control rules to prioritize acks and other <500 byte packets > over larger packets. > > And yeah, wireless is evil. Don’t trust performance tests on wireless.
I've tried all the things. And the issues happened on GbE as well. I'm mainly concerned my 6501-50 is defective in some way. With my old linux shorewall install, I had upstream QoS, prioritization of acks, ssh, other important stuff, bittorrent and other p2p went in the least important bucket, even after http/https traffic. None of it actually solved the problems I've been seeing. A combination of using my RT-AC66U instead of the soekeris (though I may try the pfsense install on the soekeris again), and switching to transmission from deluge has helped with p2p speeds (switching bt clients did not help the last few times I tried to solve this issue). I've been fighting with this for ages. And it seemed it got worse the past few months. From just having bt uploads be stupid slow and flap about (peers would connect, sit there, not get anything and drop me, or get 5KB/s and then drop me). Recently http and ssh traffic had started to get weird, even over my wired lan. Connections would take too long to start up, some websites seemed to take an age to connect, but load fast. > ED. > > > On 2015, Dec 20, at 8:14 PM, Freek Dijkstra <[email protected]> wrote: > > > > Hi Thomas, > > > > [...] > > > >> When I'm part of any kind of p2p "swarm", I can't upload reliably. It's > >> so bad that I've rarely ever seen more than 20KBps per peer, and it's > >> usually between 0 and 10KBps (5KBps is pretty average :o). > >> > >> I did some testing this weekend, and tried out the latest stable pfSense > >> on it (I was running Debian Sid), and things improved a bit. Now I'm > >> using an RT- AC66U > > > > [....] > > > > First of all, your network setup if not entirely clear. Is your p2p > > client running on your Soekris net6501? > > > > If so, is that wifi network with the Asus RT-AC66U between the Soekris > > and the upstream network? > > > > If not, why do you think the Soekris hardware is responsible for the > > poor network performance? > > > > My first two suspects of bad network performance would be: > > > > 1. Poor wifi. What is the packet loss and jitter (variation in latency) > > between the wifi client and wifi AP? Ideally, this should be about 1 ms > > RTT without much fluctuations, but 3 or 5 ms is usually acceptable too. > > If you see latency over 10 ms, or over 5% packet loss, that is most > > likely the culprit. Look into the location of the wifi, the channels > > used by you and neighbours, etc. If you haven't done so, download a wifi > > scanner to find out which channels are in use, and how strong they are. > > Note that for 2.4 GHz (802.11g, 802.11n), an AP in e.g. channel 5 may > > still interfere with channels 3 thru 7. 5 GHz (802.11n, 802.11ac) has > > less users, but is more susceptible to interference by walls, floors and > > doors. > > > > 2. Network saturation because of the use of UDP-based p2p traffic. TCP > > has a fairly commanding congestion control algorithm, which > > significantly backs-off in case of congestion. A fair portion of p2p > > clients uses UDP, which lacks congestion control, and may continue to > > send traffic. Unfortunately, this often results in higher latency, > > collisions, and ultimately collapse of performance. It certainly pays of > > to configure your p2p client to only use at max 50-75% of your upstream > > bandwidth capacity. > > > > Other sources might be poor TCP congestion control tuning, buffer bloat > > or the opposite (too small buffers), poor Ethernet flow control, poor > > upstream network. Insufficient TCP-offloading in your NIC, or -more > > generally- not powerful enough CPU in your Soekris 6501 would be at the > > bottom of list of suspects. > > > > Hope this works. > > > > Freek > > _______________________________________________ > > Soekris-tech mailing list > > [email protected] > > http://lists.soekris.com/mailman/listinfo/soekris-tech > > _______________________________________________ > Soekris-tech mailing list > [email protected] > http://lists.soekris.com/mailman/listinfo/soekris-tech -- Thomas Fjellstrom [email protected] _______________________________________________ Soekris-tech mailing list [email protected] http://lists.soekris.com/mailman/listinfo/soekris-tech
