I’m struggling with the problem that some users on one of our network-segments
get only 3Mbit/s down but 6MBit up tcp-rate. I’ve done several tests but could not see
the problem.
The Segment is behind a RB500 which is connected to a RB1xx wireless PTP 5GHz
the RB1xx is connected to the Wireless backbone via Ethernet.
Performance tests from the RB1xx gives good result, from the RB500 give
bad results. Strange: There is no performance problem using UDP. The Station on the
backbone I’m testing with (sender) is a ROS on a P4 2,8GHz.
The wireless connection is stable with good quality. Signal rate stays stable at
36MBit (configured). The cpus of Rb1xx and Rb500 are below 100%.
No queues, no filters, connection tracking disabled.
seems I’ve struggling with tcp-window size. I’ve sniffed the MT tcp-Bandwith test.
When I test over 3 wireless hops with a sum up latency of 20ms the ack’s of the
sending router aren’t fast enough to get above 5MBit/s Bandwidth.
The problem is that the receiving MT-Router has a maximal tcp-window-size
of 17376 so the sender has to wait for acks. (17376 seems to be the default
for windows, too).
If I want to achieve 6MBit/s tcp-Download the end-to-end Latency has to be
below 24ms using default max windowsize of win2000/XP/ROS.
For 3MBit/s its <48 accordingly.
This is real hard to achieve as it’s the sum of the hops within our network +
the hops through uplink to the destination. I’ve customers with 4 wireless hops.
And as our network grows …
How do you handle this? Vista seems to do some tricky things to avoid this
problem. But other windows user have to install tcpoptimizers?