LinuxCNC compatible industrial PC

More
1 year 4 months ago #329476 by PCW
Note that for Mesa Ethernet cards the latency test is not terribly useful.
What is much more important is network latency. Default Mesa driver settings
will drop (timeout) packets if the return data takes more than 80% of the
servo period. As long as you do not drop packets due to these timeouts, latency is
relatively unimportant/

if you have a mesa card, try this:

sudo chrt 99 ping -i .001 -q -c 60000 10.10.10.10

(this will run for 1 minute and print delay statistics)
 
Change the IP address if you use 192.168.1.121

 

Please Log in to join the conversation.

More
1 year 4 months ago #329557 by workshop54
Thanks PCW,

I'll give that a try.

Please Log in to join the conversation.

More
1 year 4 months ago #329559 by workshop54
Hi PCW,

Here are the results of the test you suggested. It is on a completely fresh install of LinuxCNC and I only changed the network settings using the standard ISO image. What do you think?

 

Attachment not found

Please Log in to join the conversation.

More
1 year 4 months ago #329561 by PCW
It looks like you have the host IP address set to 192.168.1.121.
This will not work, you need to set the host address to something
different but in  the same network range, say 192.168.1.100

(in other words, you were not pinging the Mesa card but rather the host itself)
 

Please Log in to join the conversation.

More
1 year 4 months ago #329673 by workshop54
I'm not sure what you mean PCW,

The ip address of the host is 192.168.1.100 as far as I can tell. (The Mesa card is connected to enp3s0)
Code:
$ ip addr 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000     link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00     inet 127.0.0.1/8 scope host lo        valid_lft forever preferred_lft forever     inet6 ::1/128 scope host noprefixroute        valid_lft forever preferred_lft forever 2: enp2s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000     link/ether 00:01:c0:33:50:2a brd ff:ff:ff:ff:ff:ff     inet 192.168.1.165/24 brd 192.168.1.255 scope global dynamic noprefixroute enp2s0        valid_lft 86236sec preferred_lft 86236sec     inet6 fe80::62f5:a9e5:df33:68d6/64 scope link noprefixroute        valid_lft forever preferred_lft forever 3: enp3s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000     link/ether 00:01:c0:33:50:2d brd ff:ff:ff:ff:ff:ff     inet 192.168.1.100/24 brd 192.168.1.255 scope global noprefixroute enp3s0        valid_lft forever preferred_lft forever     inet6 fe80::868a:991b:a25e:44d0/64 scope link noprefixroute        valid_lft forever preferred_lft forever 4: eno1: <NO-CARRIER,BROADCAST,MULTICAST,UP> mtu 1500 qdisc mq state DOWN group default qlen 1000     link/ether 00:01:c0:3a:5b:8e brd ff:ff:ff:ff:ff:ff     altname enp0s29f1 5: eno2: <NO-CARRIER,BROADCAST,MULTICAST,UP> mtu 1500 qdisc mq state DOWN group default qlen 1000     link/ether 00:01:c0:3a:5b:8f brd ff:ff:ff:ff:ff:ff     altname enp0s29f2
Code:
$ sudo chrt 99 ping -i .001 -q -c 60000 192.168.1.121 PING 192.168.1.121 (192.168.1.121) 56(84) bytes of data. --- 192.168.1.121 ping statistics --- 60000 packets transmitted, 60000 received, 0% packet loss, time 60002ms rtt min/avg/max/mdev = 0.054/0.138/0.686/0.047 ms
Code:
$ ip route get 192.168.1.121 192.168.1.121 dev enp3s0 src 192.168.1.100 uid 1000     cache

Please Log in to join the conversation.

More
1 year 4 months ago #329681 by andypugh
Late to the party, but I have recently switched both my machines to run from the Asrock N100DC-ITX board.
Latency is "fair" rather than good. But they are avaialble, run from 12V DC and can be booted from an M2 SSD for a neat installation in the machine cabinet.
www.mini-itx.com/~N100DC-ITX
(Unhelpfully, out of stock, maybe because I bought three of them from there.)
Apparently they are picky about memory, especially at 12V rather than 19V, so maybe buy memory at the same time.

Please Log in to join the conversation.

More
1 year 4 months ago #329688 by PCW
The N100DC draws no more than ~20 W...
Basically a laptop CPU in a Mini-ITX format

Please Log in to join the conversation.

More
1 year 4 months ago #329689 by PCW
I'm not sure what you mean PCW,

Your initial ping statistics had 3 and 4 usec times which are not
possible unless you are pinging the host.

Also the address range of enp3s0 overlaps enp2s0
 
This is why it's suggested to use the FPGA cards 10.10.10.10 address.
( as it will not overlap a local network DHCP range )
 

Please Log in to join the conversation.

More
1 year 4 months ago #329695 by workshop54
I tried to get the Mesa card on 10.10.10.10, but I never got that to work. I tried mesaflash --device 7i94t --addr 192.168.1.121 --set ip=10.10.10.10 before, but that didn't seem to change the boards IP address.

I ran the test again an now the values are probably more realistic:
Code:
$ sudo chrt 99 ping -i .001 -q -c 60000 192.168.1.121 PING 192.168.1.121 (192.168.1.121) 56(84) bytes of data. --- 192.168.1.121 ping statistics --- 60000 packets transmitted, 60000 received, 0% packet loss, time 59999ms rtt min/avg/max/mdev = 0.054/0.140/0.494/0.027 ms

What do you think about these values, and if they are good, are there any other tests you can recommend for this PC before I go all-in?

Please Log in to join the conversation.

More
1 year 4 months ago #329696 by PCW
Perhaps his ethernet is just really, really speedy.

Not physically possible with a 100BT link 

100 Mbits/sec is 12.5 bytes/usec = ~5 usec for a minimum size packet
since packets need to be sent and echoed that's ~10 usec absolute
minimum for the smallest  (64 byte) packet (ignoring overhead and response time)

Please Log in to join the conversation.

Time to create page: 0.137 seconds
Powered by Kunena Forum