I’ve finding the Mikrotik Router is dropping packets at times, for up to a 5 second interval, while I do a continuous ping to it from outside. I am trying to find out why it does this and thought that maybe the interface queue is backing up. Is there any way to increase this queue size? I have tons of ram
Celeron 2.8ghz, 512mb ram, Mikrotik 2.8.23
Marvel gigabit interface (LAN) (Xover to 3Com 3226 Switch, 100mbps port, FD enabled)
Intel Pro/100 (WAN) (Xover to Cox 100mbps FD ethernet feed)
12mbps - 25mbps
CPU usage 6-15%
Microsoft SMB and SNMP ports blocked
No queues except for default one
Every 30-45 minutes the unit will stop passing traffic for 4-5 seconds, then all is well for the next 30-45 minutes. It can do hours without a problem and then all of a sudden random 4-5 second packet loss over the next 8 hours.
Anyone have any thoughts? Any thoughts on how to increase the RX packet queue size for the WAN interface?
Yes, the Intel driver reports GREEN FULL DUPLEX and I can confirm with the Cox Switch its in 100mbps full duplex mode and happy. This is forced 100mbps full duplex with auto off. Tried both ways.
The Marvell/SKConnect driver for the gigabit interface is forced 100mbps full duplex and plugged into a port on the 3Com 3226 Switch. I tried a gigabit port and gigabit setting and it made no difference. Switch reports that its in 100 FD mode, Winbox reports RATE UNKNOWN and the Full Duplex light is OFF. Yet I know at 25mbps I would see alot more packet loss and collissions if it wasn’t in FD mode. Switch says FD and no Collissions so I am sure its in 100 full duplex mode.
I think this is a limitation/bug in the ‘Marvell SysKonnect SK-98xx/SK-95xx’ drivers that prevents it from properly reporting FULL DUPLEX and RATE STATUS
Well then it gets alot more difficult, as all i could think of was the auto detection renegotiating over and over (and Intel drivers auto neg. takes a few seconds on MT).
Is it narrowed down to the external interface?
It could easily be a problem on the Cox end, perhaps you should ask for an IP on the vlan you have (i assume it’s a vlan) so that you could ping that ip and see if it’s the switch.
Then i’d setup a continous ping for that IP and their router and see if it’s the first hop. Have they verified it’s not sudden bursts of traffic from some other customer?
Check for CRC errors on the switch. I have Ciscos and 3xIntel Gige and single intel 100meg Nics since i upgraded to 2.8.24 weird problems == the FD lite in winbox is off for some interfaces – no collisions but apparently at high traffic we see CRCs (not a lot) just enough for us to make out the intermittent ping loss.
Setting to auto neg then swigtching back to FD and no auto fixed it for now
This maybe a bug?
Thanks guys, well the Marvell driver definitely has a bug where it doesn’t detect RATE or FD status, that I am sure of.
Doing some PingPlotter tests from inside going out and from outside going in I find the Cox provided Switch seems to be dropping packets between tis fiber and ethernet interfaces, thus the problem!! Cox is coming out tonight to swap the switch out with another unit. Hopefully this corrects my issue. Guess the packet loss issue wasn’t attributed to the Mikrotik router! Thank god it wasn’t my fault!
Just an update for all, seems the Cox provided AT 8724XL switch couldn’t handle some packet spikes we were receiving from the internet. We would go from 4000 pps to 8-12,000 pps for a few seconds, during that time the switch would drop packets!
They replaced the switch with a higher end Cisco 3550 which seems to be doing a better job, no packet loss since the new switch was installed. Hopefully this switch can scale better as our bandwidth grows.
Make sure you check your duplex settings on Cisco switch on the ports and match them to the nics. We use the same and if you do auto neg. or have wrong duplex things will work but as you scale the bw you will start seeing a lot of slowdown due to CRC errors on the switch.