I did a multihop setup of a link towards my gateway like this:
All devices are composed of RB433AH running 5.7 & latest firmware, R52nM, dual-pol 32dBi dish. Wireless links are done as following:
bridge on one side, station pseudobridge on the other
band: 5Ghz-only-N
protocol: NV2
DFS: no radar detect
WMM: disabled
datarates: all a/b/g unchecked
Hw retries: 15
Adaptive noise imunity: ap and client
HT chains: both rx/tx
HT guard: any
HT AMPDU: 0
MCS: 6,7,8,9,10,11,12,13,14,15
WDS: disabled
NV2 TDMA period: 3ms
There is a MPLS running on all routers, with transport address on Lo interface, and as per the advices in this forum I set VPLS tunnel between R1 and R4, bridged with Eth on each side. Bridges are in RSTP protocol mode.
The distances between R1 and R2 are 8km and between R3 and R4 about 17km. When I do bw test between R1 and R2, I get average 115 / 126 Mbps full duplex UDP and about 61 Mbps per TCP. Between R3 and R4 89 / 100 Mbps fd UDP and 58 Mbps TCP. That is when tested from one rb to another with CPU slammed at 100% therefore, I trust it goes higher. However, when I do a test from R1 to R4 or from the office to the gateway at the ends of the VPLS tunnels, I get max 22 Mbps per TCP.
Does anyone see what am I doing wrong here? Please, do help!
First, move from station pseudobridge to station.
Next, mpls mtu?
Disable connection tracking on ap´s.
MCS6,7? Disable these and enable only dual chain MCS(8~15).
CCQ? Signal?
moved to station, instead of station pseudobridge;
mcs 8 - 15
Connection tracking was already turned off.
The signal, on a longer link (17 km) is -57/-56 dBi, noise floor 62 dBi and CCQ during activity, goes from 97/95% to 100/100%. On a shorter one (8 km), its -51/-51, noise floor 68 dBi and CCQ 90/90%.
MPLS MTU is set on all interfaces 1522.
However, I still get the same results. Between direct points I get good results, but from R1 to R4 I get like nothing. The traffic is running trough VPLS tunnel.
When I set the same links in WDS + Nstream, I got about 65 - 70 MBps of real traffic. I trust NV2 should give me at least 20% more, judging by the results between dirctly connected points.
Question, should I be doing VPLS from R1 to R2 and from R3 to R4 instead?
