Since my first post about this got no replies, I'll rephrase.
I am experiencing extremely high latency with my network interface. AMD X2 64, Gigabyte K8N Pro-SLI, using the built-in gigabit NIC which I believe is using the forcedeth driver.
Transfer speeds are fine, but latency is ridiculous. Does the forcedeth driver have known issues like this? Do I need to invest in different hardware?
I'm not sure how to test the performance of my NIC to try to troubleshoot why this is so bad. Can someone point me in the right direction?
Thanks, Dan
Hi
What exactly do you mean with high latency? Do you mean that it takes a long time to connect (anywhere)?
There are a few precautions to follow with gigabit ethernet.
1. Only use CAT-5 cables. 2. To obtain highest possible speed on the LAN, use a Gigabit switch/hub/router 3. Make sure speed autonegociation is turned on.
giving the blame to your nic is easy but 8 times out of ten it's not the culprit.
Other devices on your network may be transmitting (broadcasting) a lot of giberish.
Use network scan tools to find out what is happening, also check your messages log.
Do a service network restart and then tail /var/log/messages -n 20 and see at what speed it came on
Another thing you might want to check is how fast your dns server is responding. It is not unusual for "shady" providers to have quite a bit of lag on their dns servers. try pinging the ip address of one of the dns servers; look in /etc/resolv.conf for the addresses
Tools like netstat and ethereal are your friend as well.
HTH
Andy
On Saturday 29 October 2005 06:06, Dan Hensley wrote:
Since my first post about this got no replies, I'll rephrase.
I am experiencing extremely high latency with my network interface. AMD X2 64, Gigabyte K8N Pro-SLI, using the built-in gigabit NIC which I believe is using the forcedeth driver.
Transfer speeds are fine, but latency is ridiculous. Does the forcedeth driver have known issues like this? Do I need to invest in different hardware?
I'm not sure how to test the performance of my NIC to try to troubleshoot why this is so bad. Can someone point me in the right direction?
Thanks, Dan
On Sat, 2005-10-29 at 20:26 +0200, Andy Pieters wrote:
Hi
What exactly do you mean with high latency? Do you mean that it takes a long time to connect (anywhere)?
My network appears to work fine from an initial look. I can connect to the internet just fine. I achieve composite download speeds that the other machines on my network see--about 500 kb/sec (I have cable).
However, something is wrong. The easiest place for me to see this is playing bzflag. Even connecting to a server running on my machine, lagstats shows that my lag (round trip from client to server and back) is 2-4 ms, but the variability is +/- 400 msec. When I connect to a server on the internet, it's much worse. Lag of 100 ms (which is typical and not a problem), but variability of +/- 5000 ms. Which means the game is completely unplayable, even on my local network.
Absolutely nothing on my internal network has changed. My new machine is now the 2nd machine with gigabit ethernet. The other is a Windows machine (Dell), which has no problems. My switches are all 100mbit, as are the other two computers. I replaced my 5 year old Athlon 1200 with this AMD X2 machine. I plugged my new machine into the network using the exact same cable as the old machine. My old machine had no problems like this--bzflag was playable, even though my video refresh rate was not that great--but network connectivity was fine. All my other machines are fine as well in this regard.
There are a few precautions to follow with gigabit ethernet.
- Only use CAT-5 cables.
That's what I have.
- To obtain highest possible speed on the LAN, use a Gigabit
switch/hub/router
I'll get to this eventually. I'm not so concerned with the intranet speed right now. My internet latency is the problem on my new FC4 machine.
- Make sure speed autonegociation is turned on.
This is turned on on all my switches. The one my AMD X2 is connected to shows that it has negotiated 100mbit full duplex. So that all appears fine.
giving the blame to your nic is easy but 8 times out of ten it's not the culprit.
I strongly suspect either the kernel or the forcedeth driver, but I'm not sure how to determine what it is.
Other devices on your network may be transmitting (broadcasting) a lot of giberish.
I highly doubt this is the problem, unless FC4 is very sensitive to this. My Athlon 1200 was running FC3, and I had never experienced this issue with the 6 other machines I have had connected to this network.
Use network scan tools to find out what is happening, also check your messages log.
Do a service network restart and then tail /var/log/messages -n 20 and see at what speed it came on
Did that, but I got no new messages in my log. I also scanned all of the messages I got today from the time I turned on my machine, and nothing out of the ordinary.
Another thing you might want to check is how fast your dns server is responding. It is not unusual for "shady" providers to have quite a bit of lag on their dns servers. try pinging the ip address of one of the dns servers; look in /etc/resolv.conf for the addresses
Unfortunately my dns servers don't respond to ping.
Tools like netstat and ethereal are your friend as well.
I'll have to read the manpages for netstat. I've used it before, but don't know much about it. I'm really looking for some sort of comprehensive benchmarking utility. I was also hoping someone else had run into this problem, and could tell me what is happening.
Thanks, Dan
HTH
Andy
On Saturday 29 October 2005 06:06, Dan Hensley wrote:
Since my first post about this got no replies, I'll rephrase.
I am experiencing extremely high latency with my network interface. AMD X2 64, Gigabyte K8N Pro-SLI, using the built-in gigabit NIC which I believe is using the forcedeth driver.
Transfer speeds are fine, but latency is ridiculous. Does the forcedeth driver have known issues like this? Do I need to invest in different hardware?
I'm not sure how to test the performance of my NIC to try to troubleshoot why this is so bad. Can someone point me in the right direction?
Thanks, Dan
-- fedora-list mailing list fedora-list@redhat.com To unsubscribe: https://www.redhat.com/mailman/listinfo/fedora-list
Andy Pieters
Another thing you might want to check is how fast your dns server is responding. It is not unusual for "shady" providers to have quite a bit of lag on their dns servers. try pinging the ip address of one of the dns servers; look in /etc/resolv.conf for the addresses
Dan Hensley:
Unfortunately my dns servers don't respond to ping.
Try the "dig" tool on a domain name. It also reports the time taken.
I'm wondering whether you'd benefit from running your own DNS server, if your ISP's were slow? It is an advantage in my case, but you'd, first, need to determine whether this was the cause of your problems.
On Sun, 2005-10-30 at 19:16 +1030, Tim wrote:
Andy Pieters
Another thing you might want to check is how fast your dns server is responding. It is not unusual for "shady" providers to have quite a bit of lag on their dns servers. try pinging the ip address of one of the dns servers; look in /etc/resolv.conf for the addresses
Dan Hensley:
Unfortunately my dns servers don't respond to ping.
Try the "dig" tool on a domain name. It also reports the time taken.
I tested my nameservers, and the first time it took about 200 msec. Subsequent times took about 60 msec.
I'm wondering whether you'd benefit from running your own DNS server, if your ISP's were slow? It is an advantage in my case, but you'd, first, need to determine whether this was the cause of your problems.
I honestly doubt that this is the problem. I've never had to consider it before, and I've been using Comcast with these nameservers for years. The problem isn't connecting to sites or resolving their IP addresses. Once I do connect to bzflag servers, it's the high variability in the traffic to that server that is the problem. And every single server, including one on my local machine, is affected.
One potential issue is that bzflag uses UDP for its communication. It wouldn't surprise me if the forcedeth driver is focused on TCP. But I have no idea, and I'm still struggling to find a test that demonstrates this.
I found the netio package with a Linux binary, and I used it to test my machine. Throughput seems fine, but I couldn't test UDP because the port it was trying to use was already in use. I haven't gone through the source to try to fix this. And it doesn't give me the benchmark I'm looking for anyway.
I also thought I'd try using a more recent forcedeth driver than comes with the FC4 kernel, but I have not been able to find where to get it. And on top of that, the FC4 kernel build dumps errors very quickly. It's been years since I have built a kernel from scratch, but I don't remember it being particularly difficult.
Dan
-- Don't send private replies to my address, the mailbox is ignored. I read messages from the public lists.