mick crane <mick.cr...@gmail.com> writes: > On 2019-04-09 07:46, Peter Wiersig wrote: >> >> Is anything else connected to this hub? If your problems occur, is >> anything else using the hub concurrently? Can you reduce the >> connections only to server and client and maybe a internet uplink? >> Network printers can do unimaginably bad things in regard to hubs, and >> even switches.
A second time to point you to a possible problem with network printers. If you can reproduce the problem and then try to reproduce without the network printer you might have possible solution. I came across a best practice lately to isolate network printers to their own broadcast segments (VLAN/bridges) and concur. >> Can you change the hub to a real switch? >> >>> but I think it might be a hub. >>> Perhaps that is the culprit ? >> >> Perhaps. Check the "ifconfig" output on the Linux side, maybe reset the >> server, connect from windows and if you're having problems check the >> dmesg output regarding the interface. >> > how do I tell if it's a switch or a hub ? > I thought a switch sends data over the required port but a hub sends > over all the ports ? Correct, after a learning phase. > It's got "8 port switch" printed on it but if there is network activity > all the lights seem to flash. Ok, that simple you can't distinguish between hubs and switches: If you connect a new device to your network, the switch has still to ask for IP->MAC resolutions on unknown IPs, and broadcast packets need to be transmitted on all ports. But a file transfer for say a linux distribution ISO should only be flickering 2 LEDs. I found that there is a price point below which the device is no true switch anymore and should be replaced. Over here it's 50 EUR or so, for your quoted 8 ports. Correction: Probably lower at 35 EUR. Mazbe I had more ports with that other price point. > /sbin/ifconfig > enp0s25: flags=4163<UP,BROADCAST,RUNNING,MULTICAST> mtu 1500 > inet 10.0.0.3 netmask 255.255.255.0 broadcast 10.0.0.255 > inet6 fe80::219:d1ff:fe41:c769 prefixlen 64 scopeid 0x20<link> > ether 00:19:d1:41:c7:69 txqueuelen 1000 (Ethernet) > RX packets 47732498 bytes 13998322190 (13.0 GiB) > RX errors 0 dropped 5642 overruns 0 frame 0 Your errors are higher when your received packets are half of mine. But still, 5k of of 47m is a error rate of 0.01%, so a hint, but not conclusive. Does the system have any kernel messages for enp0s25 since the last boot-up? Alternatively produce new messages by disconnecting the cable for about 5 seconds and reconnect the network cable. > I've got a Buster PC with apache2, cups, dovecot, roundcube as mail and > print server. > a Buster PC I mess about on, a win 10 PC and a printer all connected to > the "switch" > the gateway is a PC with pfsense on it which is connected to "switch" > and its other network card connected to ISP router thing. > > Maybe the sluggishness I sometimes observed is Windows updating itself ? Is your network speed near your downlink speed? Do you have more than one Windows 10 machine connected to your network? I doubt that you can saturate your windows 10 network interface (probably 1 gbps) with enough bandwidth so that you could see detoriation in a ssh connection, even less possible if your devices adhere to QoS packet hints. >> For reference, here's my output: >> >> eth0: flags=4163<UP,BROADCAST,RUNNING,MULTICAST> mtu 1500 >> inet 217.172.177.159 netmask 255.255.255.0 broadcast >> 217.172.177.255 >> ether 00:19:66:f1:43:9e txqueuelen 1000 (Ethernet) >> RX packets 81994186 bytes 14753508445 (13.7 GiB) >> RX errors 14 dropped 0 overruns 14 frame 0 >> TX packets 107524155 bytes 14836289080 (13.8 GiB) >> TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0 >> >> >> Take note of the 2nd line of RX and TX status lines, there should be no >> counters there, in my case 14 overruns in regard to 819 million packets >> is a very low error rate. and indeed I misread my number and it's a magnitude lower. 81.9m as Celejar noted. but mick's error-rate is 2-3 magitudes higher. >> If in doubt verify that both sides are set to auto-negotiate and >> replace both wires from the machines to the hub with new cables. Like I said, invest in a new set of cables and see if things get better. You don't need to buy gold-plated ones, but it shouldn't be factory dropout quality either. Peter
