How to overcome in p2p networks?

Question

Cisco

Good afternoon!
The situation is quite rare, but something has become more frequent lately.
There are 2 L3 switches, between which there are trunks.
On one, int vlan is raised with a mask / 30 and on the other end of the piece of iron they ping each other with losses.
Where to start debugging?
One side:
1)
65 sup2t
s2t54-adventerprisek9-mz.SPA.122-50.SY3.bin
2) Dlink DGS3610
v10.4(3T109) Release(157885)

M9-6509#ping 10.170.1.30 repeat 100

Type escape sequence to abort.
Sending 100, 100-byte ICMP Echos to 10.170.1.30, timeout is 2 seconds:
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!.!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Success rate is 99 percent (99/100), round-trip min/avg/max = 1/2/12 ms
M9-6509#ping 10.170.1.30 repeat 100

Type escape sequence to abort.
Sending 100, 100-byte ICMP Echos to 10.170.1.30, timeout is 2 seconds:
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!.!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!.!!
Success rate is 98 percent (98/100), round-trip min/avg/max = 1/4/104 ms
M9-6509#ping 10.170.1.30 repeat 100

Type escape sequence to abort.
Sending 100, 100-byte ICMP Echos to 10.170.1.30, timeout is 2 seconds:
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!.!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Success rate is 99 percent (99/100), round-trip min/avg/max = 1/2/12 ms
M9-6509#ping 10.170.1.30 repeat 100

Type escape sequence to abort.
Sending 100, 100-byte ICMP Echos to 10.170.1.30, timeout is 2 seconds:
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!.!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!!!!!!!!!!!!!!!!!!!!!!!!!.!!!!
Success rate is 98 percent (98/100), round-trip min/avg/max = 1/2/12 ms
M9-6509#

Answer 1

sucks to test from piece of iron to piece of iron. For example, if you install cisco and huawei and run ping, the loss will be 10%. The problem is solved when setting a timeout between sending a packet. It is necessary to test, first, for example, having stupidly registered vlan in the trunk between the pieces of iron, turn it on on the computer and run iperf. And only then you can test it (but most likely it will not be necessary) and, taking into account routing, also iperfom.
Do you see degradation of services? Or is it just a simple striving for the ideal?)

Answer 2

For a start on a tsiska to look through sh int counters of errors. Further, depending on the error, look for an explanation on the tsiska website.
As an option, look at the duplex settings. The auto at a tsiska is an evil.