I'm posting this thread due to the lack of result in google solving the problem we encounter since a long time now...
We have a full mikrotik backbone with full BGP IP-transit, mikrotik switch. Customers complains sometimes that they cannot download files completely, when they are large. This is annoying for device downloading firmware or Apple pre-boot that's uses only IPV4 and do not resume.
We manage to reproduce the problem with linux clients, when using curl in IPV4 only, akamai or docker.io servers sends a TCP reset packet after a moment, stopping download. On windows, no problems.
We were thinking that's our problem came from CHR on VMWARE or ROS6 old version.
We have now started a fresh ROS7 part of our backbone with baremetal only (no more CHR) and problem persists.
Our test lab:
Linux Nuc barebone, IPV4 only, RB5009 ROS 7.9, CCR1072 (ROS 7.6) IP-transift BGP.
Downloading from IPV4 on this link:
or making a
Code: Select all
docker pull slightlytyler/large-image-test
Same config with IPV6, all runs fine even if download speed reach 1.7Gbps...
It happens also as soon as speed gains over 100 Mbps. We checked all the involved ports and we TX-drop or RX-drop counters are at zero.
Client is on 1Gbps copper, all backbone in 10Gbps. It's not related to clients because when it's connected with a SFP+ to copper adapter with direct IP-Transit on it, download runs smoothly.
We can also reproduce it with virtual linux machines connected to the backbone directly.
CCR1072 is stateless with connection tracking disabled.
Config of RB5009 is basic, with a local network and a masquerade rule to gain public IP.
Anybody encountered the same behavior ?
Best regards,