MikroTik

I've changed my ubuntu NAS to mikrotik and ran into problems:

ROS version 6.15
I have about 700 pppoe sessions, nat, 400 mbit max bandwidth.
I have double port Intel nic, ROS shows: 82576 Gigabit Network Connection (rev: 1)
Port 1 is for uplinks, 2 for downlinks, 100 vlan ifaces there.

After a several hours working I have packet loss on downlink port.

Trying to ping internal pppoe customer IP's from router and see 105 (no buffer space available).

Uplinks works as usual, cant's see any drops. Reboot helps for some time.
Downgraded to 6.7, now testing.

I suppose x86 is not the choice for new ROS versions, should migrate to CCR?

Up this topic.
ROS version 6.21
I have qos,nat,dhcp-server and some firewall rules
600 Mbit traffic
After 30 min MT drops traffic on all interfaces. Try ping any host and we see "105 (no buffer space available)". Downgrade to ROS 6.5 and problem is gone. Flow control disable. DNS cache disable.
Any ideas?

UP. 6.22 Still does not work.

Have you tried other NIC and cable? Packet loss in all versions sounds like a hardware issue

Have you tried other NIC and cable? Packet loss in all versions sounds like a hardware issue

Because of "no buffer space available" and the fact that it works in 6.7 but not in any version after that, I am guessing it is a PPP-related software issue, as I describe in the other thread (http://forum.mikrotik.com/viewtopic.php ... 21#p456298).

-- Nathan

Sorry for not informing in time, I don't remember what exactly solved that problem.
I guess that changing NIC is the way. I am still using that server, but with MT NIC.

Mikrotik do not have actual versions of igb drivers, and unfortunately they don't mind.

Now I am using RB44Ge on several NASes - no problems at all!

I had a similar issue on a CCR1036. We tried to put it into production with approximately 1700 pppoe clients, once peak hours came along, it topped off about 450mbps (normal peak for us is 650mbps) and then the packet loss started. Having to scramble to put our old router in place was no fun.

Hi,

Ran into this same issue on CHR instances in both GCP and AWS.

Do an arp ping to the gateway and the console intermittently returned: 105 (no buffer space available )
Monitor traffic on the interface and there is tons of tx-queue-drops.

Issue in both my cases was the license had not been applied and both instances were running as "free" which limits throughput to 1Mbit/s.

Added a license and all good again.