TCP over WAN performance issue

[dedvalson]

Hi all,

I am new to Centos after many years in Windows world.

I have a server running Centos 7 (3.10.0-514.el7.x86_64). It has a 10 GBit ethernet card which connects to a switch and from there to a 200 MBit WAN connection.

One of the major apps we need to run on this server is accurev, which is a version control system using TCP/IP sockets to talk to another server. The issue I am seeing is VERY poor performance over this link, on the order of 2-3 mbits / second. But the really strange thing is that I have a Windows Virtual Machine running on this same Centos 7 server. The Windows VM sees 16-20 mbits running the same test. So the VM obtains something like 8 times the network performance of the host that the VM is running on. They are both going through the same physical hardware and connection.

Since I am using VMs, I am using a bridge in Centos. The configuration for the network card is as follows:

TYPE=Ethernet
DEVICE=ens2f0
ONBOOT=yes
BRIDGE=br0

The configuration for the bridge is as follows:

TYPE="Bridge"
BOOTPROTO="none"
DEFROUTE="yes"
IPV4_FAILURE_FATAL="no"
IPV6INIT=no
NAME="br0"
DEVICE="br0"
ONBOOT="yes"
DNS1="10.101.5.20"
DNS2="10.60.8.15"
DOMAIN="global.avidww.com"
IPV6_PEERDNS="yes"
IPV6_PEERROUTES="yes"
IPV6_PRIVACY="no"
IPADDR=10.8.0.20
PREFIX=16
GATEWAY=10.8.0.2

I have tried various permutations of messing with the sysctl ipv4 memory parameters such as net.ipv4.mem, net.ipbv4.wmem, net.ipv4.tcp_window_scaling, net.ipv4.tcp_rmem, etc, but nothing seems to make much difference.

I would welcome any suggestions. It sure seems odd that the Windows VM sees so much better performance than the host. A centos 6 VM running on this server shows the same poor performance as the Centos 7 host.

Thanks,

Don