hw offloading, uplink 10Gbit. Other 1Gbit. Speed through 10Gbit to 1Gbit is under 1Gbit... about 600-700Mbit. But, 1Gbit to 1Gbit port is speed full 1Gbit. Flow control on/off... no matter.

This same problem was on CRS326. Do you test traffic at all different port speeds?

I need a solution. Immediately.

Again disappointing Mikrotik.

I have noticed some of the same problems on my CRS328-24P-4S+RM device, it does look to be related in some way to latency or packet buffers inside the switch chip.

Looking at some other posts on the forum, it appears this problem has/is with most of the CRS models. Would be nice if mikrotik could respond to this.

An iperf3 on low latency local network from 10G port to 1G port does around 930Mbit both ways, while if you put something on the 1G port that has higher latency like internet connection it can drop down to almost nothing (50-130Mbit) in the 10G to 1G direction, while the 1G to 10G direction is 930Mbit+

I think it is some kind of buffer problem on the hw level, as the problem can be solved if you can limit (queue) the 10G traffic going onto the 1G port to below 1G, then the speed goes up to full. So its likely some kind of burst problem.

I have noticed some of the same problems with my CRS328-24P-4S+RM device, it looks to be related to latency, and maybe packet buffers in the switch chip.

On the forum there are other posts of people having similar issues with other CRS models. It would be nice if mikrotik could respond to this issue.

An iperf3 on a local low latency network from 10G port to 1G port gives around 930Mbit both ways, but if something with higher latency is used like an internet connection it can drop down to almost nothing (30-130Mbit) in the 10G to 1G direction, while 1G to 10G is full speed.

It appears to be some kind of burst problem on the hw level, as the problem can be solved by limiting (queue) the traffic coming into the 10G port, that is destined to the 1G port to below 1G speeds.

First reply post didnt get though for some reason, so I try again, sorry if 2 replies show up.

i doubt flow control will fix that.
this issue is known as interface oversubscription or speed conversion drop. you fill the outgoing interface’s buffers with bursty traffic. happens if you have 10g ingress and 1g egress connection, or if multiple GE ports receive traffic that must leave on a single GE interface. the solution is usually having bigger buffers on the egress ports which can hold more packets but this causes increased latency.

Nonsense. I'll wait for someone who knows what he's talking about. There is a problem with the transition from 1G to 100M to the last CRS. The same goes for 10G to 1G.

Just have a 10Gbit uplink and only one 1Gbit device and the traffic fluctuates and 1Gbit can not be reached. At nomikrotik devices it works normally.

At nomikrotik devices it works normally.

be my guest
https://www.cisco.com/c/en/us/support/d ... ut-dr.html

nice, but the problem arises even when the 800Mbit queue is used on the uplink router to crs where the buffers in the CRS should work at rest so it does not work anyway. It still can not reach even 800Mbit. Similarly, it can be converted to 1G to 100M ports. This normally handles other devices. CRS no.

Is there any new news ? with 6.43.8