RB110Ahx2 RS v6.6: slow pppoe-performance (basicaly ftp)

Hello,

we are experiencing a very strange behaviour in our RB1100AHx2 with RouterOS v6.6. The RB is located at our datacenter with a 1GBit-connection to the internet. Our customers at different locations are connected with glasfibre to the datacenter. Each location with its own VLans (often more than one). The RB has the task to offer a pppoe-server (for 10 users) and to route some vlans to a dedicated hotspot-server.
The Problem now is: We have some customers that ordered a 100MBit/100MBit-connection via pppoe. When we test the connection we can reach download-speeds about 80-90MBit/s with http-speedtest. But when we download from a ftp-server we only can reach 20MBit/s. A ftp-upload is almost as fast as expected with about 65MBit/s.
Today i have done a lot of testing regarding mtu/mru/mss but i can’t improve the ftp-speed.
When i connect via pptp or to the Mikrotik-Hotspot-System everything is as fast as it should. So it has to be a pppoe-issue. As pppoe is not a big mystery, i am totally clueless what else it could be.
The configuration so far:
PPPoE-Server that runs on a network-bridge. The vlans are connected to that bridge. The pppoe-server ist set to standard mtu/mru with 1480. Profile has set limits to 100M/100M, use encryption yes, change mss yes.
Today i have tested:
MTU/MRU 1492, 1450, 1380. Everytime with “change mss” yes and no. When “change mss” set to no. I activated an own mangle rule like the following with 44bytes less then the mtu and 100bytes less then the mtu.

/ ip firewall mangle
add chain=forward protocol=tcp tcp-flags=syn action=change-mss tcp-mss=!0-1448 new-mss=1448

Does anyone have some tipps what i did not have in mind till now?


Thanks in advance

Nirnath

hi
do you use queues automatically set by ppp profile or radius ?
if yes -in the default used “default-small” set “Queue Size” to 50 packets or even higher

Thank you, that’s it. The solution was far to easy to even think about it…

Sorry to bother you guys again.

I was searching for a “good practice” size of the simple queue but could not find a hint. Testing has shown that about 100 should be good. Is there any other limitation to queue-size than the avaible memory on the RB? As system-> ressources states, we have more than 1GByte free memory.

i never tried - but i will check also …
but now i am off for skiing…

I changed queue-size back to 50. 100/100MBit seems to be working great with these values.

Have fun skiing. I have to wait about 4 weeks. Hope there will be enough snow in austria then. :wink: