[opensuse] How to fix network interface overruns?
I'm testing a new fibre connection at 50/10Mbit, and the network interface card is showing an increasing number of overruns when I "stress" the link with a lengthy download from a high-speed source. In the last 4Gb download, I saw the counter increase by at least 10000. Googling has informed me that "Receive overruns means that packets are arriving faster than you can receive". The firewall to which the fibre is connected is an elderly PIII 800MHz, but I can't quite imagine that causing this issue. It has four network interfaces: 02:04.0 Ethernet controller: 3Com Corporation 3c905B 100BaseTX [Cyclone] (rev 24) 02:08.0 Ethernet controller: Intel Corporation 82801BA/BAM/CA/CAM Ethernet Controller (rev 01) 02:09.0 Ethernet controller: 3Com Corporation 3c905B 100BaseTX [Cyclone] (rev 24) 02:0a.0 Ethernet controller: 3Com Corporation 3c905C-TX/TX-M [Tornado] (rev 74) on separate interrupts: (from /proc/interrupts) 16: 15819304 IO-APIC-fasteoi eth2 18: 13879018 IO-APIC-fasteoi eth3 20: 73917 IO-APIC-fasteoi eth0 21: 1667206 IO-APIC-fasteoi eth1 The problematic interface is eth3, a 3Com 3c905b. I'm contemplating replacing it with a newer intel e100 card, but I'm grasping at straws. Any other suggestions for troubleshooting this? -- Per Jessen, Zürich (11.4°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
On 4/1/2012 9:58 AM, Per Jessen wrote:
The problematic interface is eth3, a 3Com 3c905b.
I could have almost guessed that. Do you have ECN turned on? Gigabit nics are like 10 bucks, or 7.5 euros. Don't waste your time messing with it. I suppose its built onto the mother board and this is going to cost you a slot? -- _____________________________________ ---This space for rent--- -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
John Andersen wrote:
On 4/1/2012 9:58 AM, Per Jessen wrote:
The problematic interface is eth3, a 3Com 3c905b.
I could have almost guessed that.
Why? Is that card particularly problematic? I'm not emotionally attached to it, but its two cousins (905B&C) have been doing fine for a few years.
Do you have ECN turned on?
Dunno, but not that I know off.
Gigabit nics are like 10 bucks, or 7.5 euros.
Yeah, but there is no indication that anyone of them would handle 50Mbit any better than a 100Mbit card.
Don't waste your time messing with it. I suppose its built onto the mother board and this is going to cost you a slot?
Nope, the one e100 interface is built-in, the three others are plug-in cards. I've only got one 32bit PCI bus for them though. -- Per Jessen, Zürich (5.8°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
On 4/1/2012 2:24 PM, Per Jessen wrote:
John Andersen wrote:
On 4/1/2012 9:58 AM, Per Jessen wrote:
The problematic interface is eth3, a 3Com 3c905b.
I could have almost guessed that. Why? Is that card particularly problematic? I'm not emotionally attached to it, but its two cousins (905B&C) have been doing fine for a few years.
The entire line was flaky in my experience, and prone to sudden recalcitrance. Sometimes these mysteriously fix themselves with a total POWER OFF restart. The problems centered in its auto-negotiation (or lack there of) as I recall. When we were managing a large number of machines under contract, we would replace these with a 10 dollar nic at the first sign of hard to diagnose networking problems and the problems always instantly disappeared. Affected both windows and linux. Google finds a gazillion hits. -- _____________________________________ ---This space for rent--- -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
On Sun, 2012-04-01 at 18:58 +0200, Per Jessen wrote:
I'm testing a new fibre connection at 50/10Mbit, and the network interface card is showing an increasing number of overruns when I "stress" the link with a lengthy download from a high-speed source. In the last 4Gb download, I saw the counter increase by at least 10000.
Googling has informed me that "Receive overruns means that packets are arriving faster than you can receive". The firewall to which the fibre is connected is an elderly PIII 800MHz, but I can't quite imagine that causing this issue. It has four network interfaces:
02:04.0 Ethernet controller: 3Com Corporation 3c905B 100BaseTX [Cyclone] (rev 24) 02:08.0 Ethernet controller: Intel Corporation 82801BA/BAM/CA/CAM Ethernet Controller (rev 01) 02:09.0 Ethernet controller: 3Com Corporation 3c905B 100BaseTX [Cyclone] (rev 24) 02:0a.0 Ethernet controller: 3Com Corporation 3c905C-TX/TX-M [Tornado] (rev 74)
on separate interrupts: (from /proc/interrupts)
16: 15819304 IO-APIC-fasteoi eth2 18: 13879018 IO-APIC-fasteoi eth3 20: 73917 IO-APIC-fasteoi eth0 21: 1667206 IO-APIC-fasteoi eth1
The problematic interface is eth3, a 3Com 3c905b. I'm contemplating replacing it with a newer intel e100 card, but I'm grasping at straws. Any other suggestions for troubleshooting this?
Hi, Replacing it with a different nic might do the trick. John already suggeted a Gb-nic, but it's not the speed that might do the trick for you, modern nics (capable of realy doing Gb) often have longer hareware fifo's onboard, and generate less IRQ's for the same amount of data, compared with 3com's. Might consider a single Intel board with quad E1000 if irq's or the bus is a problem. hw -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
Hans Witvliet wrote:
Replacing it with a different nic might do the trick.
I'll try that anyway.
John already suggeted a Gb-nic, but it's not the speed that might do the trick for you, modern nics (capable of realy doing Gb) often have longer hareware fifo's onboard, and generate less IRQ's for the same amount of data, compared with 3com's. Might consider a single Intel board with quad E1000 if irq's or the bus is a problem.
I've also got some fairly ancient Sun HMEs with 4x100Mbit, but they're 64bit PCI cards. I still might experiment. Does anyone have any real suggestion as to why a 100Mbit ethernet card cannot handle 50Mbit? It doesn't matter much, but it doesn't seem to make much sense? -- Per Jessen, Zürich (5.1°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
Per Jessen wrote:
Does anyone have any real suggestion as to why a 100Mbit ethernet card cannot handle 50Mbit? It doesn't matter much, but it doesn't seem to make much sense?
Perhaps it's flakey, to use the technical term. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
On Sun, 2012-04-01 at 23:29 +0200, Per Jessen wrote:
Hans Witvliet wrote:
Replacing it with a different nic might do the trick.
I'll try that anyway.
John already suggeted a Gb-nic, but it's not the speed that might do the trick for you, modern nics (capable of realy doing Gb) often have longer hareware fifo's onboard, and generate less IRQ's for the same amount of data, compared with 3com's. Might consider a single Intel board with quad E1000 if irq's or the bus is a problem.
I've also got some fairly ancient Sun HMEs with 4x100Mbit, but they're 64bit PCI cards. I still might experiment.
Does anyone have any real suggestion as to why a 100Mbit ethernet card cannot handle 50Mbit? It doesn't matter much, but it doesn't seem to make much sense?
Per, Don't know if possible, but can you exclude any other influences? Normally (indeed...) it is not difficult to compleely saturate a 100Mb card, a bit harder to it for a 1Gb card, and it takes some effort to saturate a 10Gb card. Besides the droppings (shouldn't happen), how did you measure the throughput? As to the 'why question", i've witnesed mobo's, mem and nic's gone bad long after an ESD-event. And i'm not talking about lightning, but simply unsafe removing/installing components in a pc. Dutch telco KPN did some research on their own people and corelated it with failure events. After they implemented their strict "clean" policy the number of unexplained hw-failures dropped significantly. In their instruction-movie, they shown pictures from microscopic reseach on chips and pcb's. Something i'll never forget. hw -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
Hans Witvliet wrote:
On Sun, 2012-04-01 at 23:29 +0200, Per Jessen wrote:
Does anyone have any real suggestion as to why a 100Mbit ethernet card cannot handle 50Mbit? It doesn't matter much, but it doesn't seem to make much sense?
Per, Don't know if possible, but can you exclude any other influences?
Not really - the new link is hooked up to the firewall which has existing traffic.
Normally (indeed...) it is not difficult to compleely saturate a 100Mb card, a bit harder to it for a 1Gb card, and it takes some effort to saturate a 10Gb card.
Besides the droppings (shouldn't happen), how did you measure the throughput?
Completely unscientifically - I ran wget on a large file, but with --limit-rate. Every minute I checked the overruns: 100000 bytes/sec – no overruns 500000 bytes/sec – very few. Minutes without any. 1000000 bytes/sec – 10/minute, 19/minute, 11/minute. 2000000 bytes/sec – 130/minute, 130/minute, 130/minute no limit – 3-400 per minute.
As to the 'why question", i've witnesed mobo's, mem and nic's gone bad long after an ESD-event. And i'm not talking about lightning, but simply unsafe removing/installing components in a pc.
Dutch telco KPN did some research on their own people and corelated it with failure events. After they implemented their strict "clean" policy the number of unexplained hw-failures dropped significantly. In their instruction-movie, they shown pictures from microscopic reseach on chips and pcb's. Something i'll never forget.
Yes, the card could be flakey, I'll try with another one tonight. -- Per Jessen, Zürich (6.7°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
Per Jessen wrote:
Hans Witvliet wrote:
On Sun, 2012-04-01 at 23:29 +0200, Per Jessen wrote:
Does anyone have any real suggestion as to why a 100Mbit ethernet card cannot handle 50Mbit? It doesn't matter much, but it doesn't seem to make much sense?
Per, Don't know if possible, but can you exclude any other influences?
Not really - the new link is hooked up to the firewall which has existing traffic.
Normally (indeed...) it is not difficult to compleely saturate a 100Mb card, a bit harder to it for a 1Gb card, and it takes some effort to saturate a 10Gb card.
Besides the droppings (shouldn't happen), how did you measure the throughput?
Completely unscientifically - I ran wget on a large file, but with --limit-rate. Every minute I checked the overruns:
100000 bytes/sec – no overruns 500000 bytes/sec – very few. Minutes without any. 1000000 bytes/sec – 10/minute, 19/minute, 11/minute. 2000000 bytes/sec – 130/minute, 130/minute, 130/minute no limit – 3-400 per minute.
As to the 'why question", i've witnesed mobo's, mem and nic's gone bad long after an ESD-event. And i'm not talking about lightning, but simply unsafe removing/installing components in a pc.
Dutch telco KPN did some research on their own people and corelated it with failure events. After they implemented their strict "clean" policy the number of unexplained hw-failures dropped significantly. In their instruction-movie, they shown pictures from microscopic reseach on chips and pcb's. Something i'll never forget.
Yes, the card could be flakey, I'll try with another one tonight.
Okay, have just now replaced the 3c905B with a 3c905C - I'm still getting overruns: Single highspeed download source - 100/minute. Multiple concurrent downloads - 600-700/minute I'll try out an intel e100 card later, but I suspect the cards are all fine. -- Per Jessen, Zürich (16.4°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
On Mon, 2012-04-02 at 19:18 +0200, Per Jessen wrote:
Okay, have just now replaced the 3c905B with a 3c905C - I'm still getting overruns:
Single highspeed download source - 100/minute. Multiple concurrent downloads - 600-700/minute
I'll try out an intel e100 card later, but I suspect the cards are all fine.
Ok Per, your logic seems solid... So it appears that your (3C905b) board isn't defect. However, i presume there isn't much difference between the "-B" and the "-C" version. So if you move towards a completely different vendor (different size of internal FIFO perhaps), like Intel, you still might see a difference. But could it be that your cpu can not handle that amount of IRQ's? (Wild guess, as 50Mbps isn't _that_ much ...) Hans -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
Hans Witvliet wrote:
On Mon, 2012-04-02 at 19:18 +0200, Per Jessen wrote:
Okay, have just now replaced the 3c905B with a 3c905C - I'm still getting overruns:
Single highspeed download source - 100/minute. Multiple concurrent downloads - 600-700/minute
I'll try out an intel e100 card later, but I suspect the cards are all fine.
Ok Per, your logic seems solid...
So it appears that your (3C905b) board isn't defect. However, i presume there isn't much difference between the "-B" and the "-C" version.
One is called "Cyclone", the other one "Tornado". They're handled by the same driver.
So if you move towards a completely different vendor (different size of internal FIFO perhaps), like Intel, you still might see a difference.
Yes, that is the next step. I did replace the 3c905c with a tlan last night, but I couldn't get to test it.
But could it be that your cpu can not handle that amount of IRQ's? (Wild guess, as 50Mbps isn't _that_ much ...)
Yes, that was also my first thought, but it doesn't make much sense. Does anyone know how to determine if that could be the reason? Just to eliminate that suspect. -- Per Jessen, Zürich (10.9°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
Per Jessen wrote:
Per Jessen wrote:
Okay, have just now replaced the 3c905B with a 3c905C - I'm still getting overruns:
Single highspeed download source - 100/minute. Multiple concurrent downloads - 600-700/minute
I'll try out an intel e100 card later, but I suspect the cards are all fine.
Okay, I'm now running with an Intel e100 card, and it's looking a lot better: Single highspeed download source - 1-2 overruns per minute. (this source can drive the link at sustained 3.5-4Mbyte/s, peaks at up to 5-6Mbyte/s). Multiple concurrent downloads - 4-5 overruns per minute. I can probably live with this, but it's annoying. -- Per Jessen, Zürich (16.1°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
On Tue, 3 Apr 2012 23:20:57 Per Jessen wrote:
Per Jessen wrote:
Per Jessen wrote:
Okay, have just now replaced the 3c905B with a 3c905C - I'm still getting overruns:
Single highspeed download source - 100/minute. Multiple concurrent downloads - 600-700/minute
I'll try out an intel e100 card later, but I suspect the cards are all fine.
Okay, I'm now running with an Intel e100 card, and it's looking a lot better:
Single highspeed download source - 1-2 overruns per minute. (this source can drive the link at sustained 3.5-4Mbyte/s, peaks at up to 5-6Mbyte/s).
Multiple concurrent downloads - 4-5 overruns per minute.
I can probably live with this, but it's annoying.
TX or RX overruns? They're usually caused by duplexing errors, maybe due to incorrect auto-negotiation of link parameters, but can also be caused by faulty or out-of-spec cabling. Have you properly tested your patch and/or structured cabling plant? I'm not talking about just continuity/wire-map testing, but properly testing to certified standards including cross-talk, insertion loss etc? Also, try locking down one or both ends of the network link to a fixed speed/duplex setting (e.g. both ends at 100Mb FD. If that doesn't help, try dropping it to a lower speed/duplex setting and see it it helps. Regards, Rodney. -- ========================================================================== Rodney Baker VK5ZTV rodney.baker@iinet.net.au ========================================================================== -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
Rodney Baker wrote:
On Tue, 3 Apr 2012 23:20:57 Per Jessen wrote:
Per Jessen wrote:
Per Jessen wrote:
Okay, have just now replaced the 3c905B with a 3c905C - I'm still getting overruns:
Single highspeed download source - 100/minute. Multiple concurrent downloads - 600-700/minute
I'll try out an intel e100 card later, but I suspect the cards are all fine.
Okay, I'm now running with an Intel e100 card, and it's looking a lot better:
Single highspeed download source - 1-2 overruns per minute. (this source can drive the link at sustained 3.5-4Mbyte/s, peaks at up to 5-6Mbyte/s).
Multiple concurrent downloads - 4-5 overruns per minute.
I can probably live with this, but it's annoying.
TX or RX overruns?
Only RX overruns.
They're usually caused by duplexing errors, maybe due to incorrect auto-negotiation of link parameters, but can also be caused by faulty or out-of-spec cabling.
Have you properly tested your patch and/or structured cabling plant? I'm not talking about just continuity/wire-map testing, but properly testing to certified standards including cross-talk, insertion loss etc?
No, nothing like that. Haven't got the equipment for that.
Also, try locking down one or both ends of the network link to a fixed speed/duplex setting (e.g. both ends at 100Mb FD.
Yes, they're both at 100Mbit FD.
If that doesn't help, try dropping it to a lower speed/duplex setting and see it it helps.
Well, I need to push 50Mbit/s downstream ... -- Per Jessen, Zürich (18.4°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
On 4/3/2012 7:21 AM, Rodney Baker wrote:
TX or RX overruns? They're usually caused by duplexing errors, maybe due to incorrect auto-negotiation of link parameters, but can also be caused by faulty or out-of-spec cabling.
I can virtually assure you that is is an auto-negotiation error with the card Per was using. This entire card line has a history of auto-negotiation failures. Often they will work for months with no problem, then they simply fail to negotiate a fast connection, decide to fix themselves in 10meg mode regardless of what the driver asks for, or fail to honor full duplex, of something. Often you can force them using mii-tool, or something, but it will come back to bite you after an upgrade, or after you cable them to a new switch. Sometimes powering them complete off (pull-the-plug off, not simply power-down-off due to wake on lan capabilities) they will reset and work ok. (Only to strike again later). I just replace them at the first sign of trouble. -- _____________________________________ ---This space for rent--- -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
John Andersen wrote:
On 4/3/2012 7:21 AM, Rodney Baker wrote:
TX or RX overruns? They're usually caused by duplexing errors, maybe due to incorrect auto-negotiation of link parameters, but can also be caused by faulty or out-of-spec cabling.
I can virtually assure you that is is an auto-negotiation error with the card Per was using. This entire card line has a history of auto-negotiation failures.
Still, both the 3c905b and the 3c905c showed poor performance. (=lots of overruns). The Intel e100 that is now running is also showing overruns, although a lot less. I'm not at all convinced this is a hardware issue, but I am curious about what will happen when we up the line to 100Mbit/s. (in a year or two). -- Per Jessen, Zürich (10.6°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
Hans Witvliet wrote:
As to the 'why question", i've witnesed mobo's, mem and nic's gone bad long after an ESD-event. And i'm not talking about lightning, but simply unsafe removing/installing components in a pc.
FWIW, I recently had a NIC go bad. It hadn't been removed & reinstalled since I first bought it a few years ago. It would work fine and then simply stop working for a few minutes, then start working again. So, a defective NIC is certainly a possibility. -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
* Per Jessen <per@computer.org> [04-01-12 13:11]:
I'm testing a new fibre connection at 50/10Mbit, and the network interface card is showing an increasing number of overruns when I "stress" the link with a lengthy download from a high-speed source. In the last 4Gb download, I saw the counter increase by at least 10000.
Googling has informed me that "Receive overruns means that packets are arriving faster than you can receive". The firewall to which the fibre is connected is an elderly PIII 800MHz, but I can't quite imagine that causing this issue. It has four network interfaces:
When you get it solved, I would be interested. I have been getting: Info: Traffic rate for "eth0" higher than set maximum 100 Mbit (54->743, r199 t3504), syncing for some time now. hwinfo --netcard 27: PCI 107.0: 0200 Ethernet controller [Created at pci.318] UDI: /org/freedesktop/Hal/devices/pci_1317_985 Unique ID: rBUF.IH1TcNNtXe0 Parent ID: WL76.vuIf6LvchL4 SysFS ID: /devices/pci0000:00/0000:00:09.0/0000:01:07.0 SysFS BusID: 0000:01:07.0 Hardware Class: network Model: "ADMtek NC100 Network Everywhere Fast Ethernet 10/100" Vendor: pci 0x1317 "ADMtek" Device: pci 0x0985 "NC100 Network Everywhere Fast Ethernet 10/100" SubVendor: pci 0x1317 "ADMtek" SubDevice: pci 0x0574 Revision: 0x11 Driver: "tulip" Driver Modules: "tulip" Device File: eth0 I/O Ports: 0xac00-0xacff (rw) Memory Range: 0xfdfff000-0xfdfff3ff (rw,non-prefetchable) Memory Range: 0xfde00000-0xfde1ffff (ro,non-prefetchable,disabled) IRQ: 17 (3700794 events) HW Address: 00:04:5a:55:39:c8 Link detected: yes Module Alias: "pci:v00001317d00000985sv00001317sd00000574bc02sc00i00" Driver Info #0: Driver Status: tulip is active Driver Activation Cmd: "modprobe tulip" Config Status: cfg=no, avail=yes, need=no, active=unknown Attached to: #18 (PCI bridge) -- (paka)Patrick Shanahan Plainfield, Indiana, USA HOG # US1244711 http://wahoo.no-ip.org Photo Album: http://wahoo.no-ip.org/gallery2 http://en.opensuse.org openSUSE Community Member Registered Linux User #207535 @ http://linuxcounter.net -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
El 01/04/12 13:58, Per Jessen escribió:
I'm testing a new fibre connection at 50/10Mbit, and the network interface card is showing an increasing number of overruns when I "stress" the link with a lengthy download from a high-speed source. In the last 4Gb download, I saw the counter increase by at least 10000.
Googling has informed me that "Receive overruns means that packets are arriving faster than you can receive". The firewall to which the fibre is connected is an elderly PIII 800MHz, but I can't quite imagine that causing this issue. It has four network interfaces:
02:04.0 Ethernet controller: 3Com Corporation 3c905B 100BaseTX [Cyclone] (rev 24) 02:08.0 Ethernet controller: Intel Corporation 82801BA/BAM/CA/CAM Ethernet Controller (rev 01) 02:09.0 Ethernet controller: 3Com Corporation 3c905B 100BaseTX [Cyclone] (rev 24) 02:0a.0 Ethernet controller: 3Com Corporation 3c905C-TX/TX-M [Tornado] (rev 74)
on separate interrupts: (from /proc/interrupts)
16: 15819304 IO-APIC-fasteoi eth2 18: 13879018 IO-APIC-fasteoi eth3 20: 73917 IO-APIC-fasteoi eth0 21: 1667206 IO-APIC-fasteoi eth1
The problematic interface is eth3, a 3Com 3c905b. I'm contemplating replacing it with a newer intel e100 card, but I'm grasping at straws. Any other suggestions for troubleshooting this?
Hi, did you tried RPS/RFS ? http://code.google.com/p/kernel/wiki/NetScalingGuide#RPS_Configuration -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
Cristian Rodríguez wrote:
El 01/04/12 13:58, Per Jessen escribió:
I'm testing a new fibre connection at 50/10Mbit, and the network interface card is showing an increasing number of overruns when I "stress" the link with a lengthy download from a high-speed source. In the last 4Gb download, I saw the counter increase by at least 10000.
Googling has informed me that "Receive overruns means that packets are arriving faster than you can receive". The firewall to which the fibre is connected is an elderly PIII 800MHz, but I can't quite imagine that causing this issue. It has four network interfaces:
02:04.0 Ethernet controller: 3Com Corporation 3c905B 100BaseTX [Cyclone] (rev 24) 02:08.0 Ethernet controller: Intel Corporation 82801BA/BAM/CA/CAM Ethernet Controller (rev 01) 02:09.0 Ethernet controller: 3Com Corporation 3c905B 100BaseTX [Cyclone] (rev 24) 02:0a.0 Ethernet controller: 3Com Corporation 3c905C-TX/TX-M [Tornado] (rev 74)
on separate interrupts: (from /proc/interrupts)
16: 15819304 IO-APIC-fasteoi eth2 18: 13879018 IO-APIC-fasteoi eth3 20: 73917 IO-APIC-fasteoi eth0 21: 1667206 IO-APIC-fasteoi eth1
The problematic interface is eth3, a 3Com 3c905b. I'm contemplating replacing it with a newer intel e100 card, but I'm grasping at straws. Any other suggestions for troubleshooting this?
Hi, did you tried RPS/RFS ?
http://code.google.com/p/kernel/wiki/NetScalingGuide#RPS_Configuration
Thanks for the link, that was interesting reading. My firewall only has only one elderly CPU though :-) -- Per Jessen, Zürich (16.9°C) -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse+owner@opensuse.org
participants (7)
-
Cristian Rodríguez -
Hans Witvliet -
James Knott -
John Andersen -
Patrick Shanahan -
Per Jessen -
Rodney Baker