What is the problem with the windows\linux\cisco link?

Question

linux

The situation is such
an OFFICE and a WAREHOUSE
OFFICE: cisco 2921, 2 providers work, KERIO WAREHOUSE
: cisco 1921 (no shaper)
Recently, on the main channel with a warehouse, the channel fell by exactly half (planned "measurements" by iperf'om).
Here are the measurements on the main provider: (all through iperf) and comparison
Windows SKLAD (server) - windows OFFICE (client) - 8-9mbits / s
Windows SKLAD (client) - windows OFFICE (server) -8-9mbits / s
Windows SKLAD ( server) – centos OFFICE(client) – 17-20mbits/s
Windows SKLAD (client) – centos (server) – 8-9 mbits/s
Centos SKLAD (server) – centos OFFICE (client) 18-19mbits/s
Centos SKLAD ( client) – centos OFFICE (server) 17-19mbits/s
tcpdump outputs on centos: (IPERF)
13: 22: 00.970290 IP 192.168.20.4.40794> 192.168.99.16.commplex-link: flags [.], SEQ 1441816: 1443176, ACK 1, Win 515, Length 1360
Wireshark (Windows)
33 0.122191000 192.168.99.213 TCP 1414 40786 > commplex-link [ACK] Seq=38233 Ack=1 Win=131840 Len=1360
It differs mainly in the size of the transmission window (on windows it is noticeably larger). I don’t understand batch analysis well, that’s why I ask for help, why such a strange channel? centos-machines skip the full channel, and with windows - halved
* antiviruses, firewalls, etc. - disabled
** tried directly through providers (bypassing the office gateway) - the same
*** cisco configs are excellent and worked out over the years

Answer 1

the problem was the tcp window size. The increase corrected the situation.