Hello!

I discovered a serious problem with 750G switch

Reproductible scenario:
2 pcs 750G ROS4.16
many 433AH ROS4.16
1 RB600 ROS4.16

Connect 750G to another, and connect 433AH to any port on any device and connect RB600 to any 750G

Using wire speed switching via switch chip.

Each 433AH connect 100Mbit/s with switch, and could transmit and receive full 100Mbit traffic with another 433AH, tested with bandwidth test, and real world (ftp, web) traffic
this is expected, and fine

RB600 connect to 750G via 1Gbit/s and measured 300-400Mbit traffic (measured with bandwidth-test and this is CPU limit of the test would be higher if not making on the transmitting device) this is fine.

If measure from the another 750G speed also 300-400Mbit (this is fine)

Problem:
If i measure traffic between any 100Mbit connected device to ANY (750G or RB600) device the traffic (receive traffic from Gbit connected device) only 30Mbit/s not more this is very bad because this device should used to uplink gbit traffic from 100Mbit devices, but now lower than 100Mbit if i want use this function. (transmit traffic not affected from 100Mbit)

Short description: 750G cannot make over 30Mbit/s traffic between connected 1Gbit and 100Mbit interface if traffic direction Gbit->100Mbit via switch chip

The problem may with backpressure logic in switch chip between Gbit/100Mbit ports, because if i set Gbit interface to 100Mbit speed(force) the full of 100Mbit traffic achieved on this port also. (the problem not related to other cited device in this example)

Mikrotik, please fix this problem asap thanks!

这是问题实际auto-n吗egotiation enabled on both ends? Please make sure that this is not the case ofhttp://en.wikipedia.org/wiki/Duplex_mismatch.

My previous post may not understandable enough. Many test and hour behind the post above.
(The short description very clear, please verify)

Please make a testbed config same devices, and you see: this is not duplexity problem.
If would duplex mismatch the problem affect all port regardless of connection speed and traffic direction

This problem affected only if you use 1 port in giga mode and another 100Mbit and make traffic between these ports.
If you make traffic from gbit port the overall speed only 30Mbit/s. If i limit gbit port to 100Mbit works 100Mbit fully.
If i test not receive speed (gbit/s->100mbt/s) but the transmit speed (100Mbit/s->1gbit/s) the speed is normal 100Mbit/s
Also same time if make traffic between trought any 100Mbit/s port (regardless off traffic direction) the device.

Problem may related to backpressure (because only affect speed if traffic come from gbit -> 100mbit/s interface)

FYI Duplexity: all ports auto-negotiated, and detect full duplex all devices all ports, in normal test case.
(and remember this is a 750G with statistics on ethernet devices and i probably not make any statement if i see any error on any interface)

Just to test it I turned off auto negotiation on one port of a RB750G and fixed it to 100 Mbps full duplex. The other side was a PC 1Gbps card with auto negotiation enabled. I got dramatic drop of actual data transfer speed from 70 Mbps to about 0.5 Mbps.

Just to test it I turned off auto negotiation on one port of a RB750G and fixed it to 100 Mbps full duplex. The other side was a PC 1Gbps card with auto negotiation enabled. I got dramatic drop of actual data transfer speed from 70 Mbps to about 0.5 Mbps.

Please check the link about duplex mismatch.

Ivoshiee

Please do not offtopic this thread, this problem _not_ related to duplex problem, and wrong setting may cause problems on your boxes (please don't check off auto negotioation on your boxes only if this is the problem, and in this situation, make proper settings at the other end).

Thanks!

Post all router configs please, We have a large number of RB750G deployed including setups where 1 port links at gbit and the other at 100mbit and dont see your problem. Also is the cabling made by you or precrimped named brand cables? We have had a number of gbit links act funny when using handmade cables under load

Duplex/AutoNeg bugs do give extremely odd behavior depending on the chips involved in the link

Problem only appear if you contact 100mbit device trought 1Gbit 750G interface, and you use the switch chip.

750G (A device)
750克(B设备)gydF4y2Ba
433AH (C device)

750G connect with 1 Gbit/s use original patch utp 2 meters (on desk)

connect 433AH to any 750G

Ok choose device B to connect C device (A and B devices are 5 port switch (hw) nothing special after reset only an IP set on the master port)
if you make bw test from A device(750G) to C device (433AH) in transmit speed only 28-30Mbit/s (to 433AH ( measured with tool bandwidth-test CIPADDRES direction=transmit on A device) but 97Mbit/s (from 433 AH (direction=receive (any other same) (750G switch chip make backpressure from 1Gbit traffic to 100Mbit traffic but fail)
If you make bw test from A device(750G) to B device (750G second) transmit and receive also 500Mbit/s (the cable good, and device works fine.
If you make bw test from B device(750G) to C device (433AH) transmit and receive also 97Mbit/s (the cable good, and device works fine.

No errors on devices (interface ethernet print stats) no duplex problem no cable problem tested many hours many config many devices.

Problem: i cant make aggregation of 100Mbit/s devices because 750G fail (regadless of config)