I would greatly appreciate guidance in optimizing a point to point link. Please forgive me if I've made obvious newbie mistakes, I first touched RouterOS less than a year ago.

I have 2x SXT 2nD r2 configured for a bridged point to point link. 
Distance: just under 1k feet. (300 meters).

My client is now attempting to run an old foxpro database application across this link. The software is not optimized, but there's nothing I can do on that end. Latency is causing performance issues (2 min vs 15 min to run a report I'm testing with). It was taking 25 minutes to run this report from remote network, before I identified noise that was not present on install and changed the frequency of the APs. The performance increase directly correlates to measured improvement in latency.

Using ping as a diagnostic tool, response time is 2-3 ms, with regular spikes to 20 ms. Average is 5 ms. Througput is roughly 60/60 Mbps.

Configs from both APs are attached.

Further testing: RB ping tool AP1 to AP2, min 1 ms, avg 7 ms, max 98 ms

With those results, latency can't be your issue.

Database queries should he fine with averages of 30ms... You're way below that...

Another persective: VoIP depends on latency way more than your DB app and you can easily make calls with what you're showing us.

Yes, you are correct, modern databases that are cleint/server oriented work fine on that connection. However, this is foxpro accessing files over SMB. SMB 1 at that. (Don't ask). 

A low latency P2P link is the only option, aside from approximately $10k in RDS licensing.


With further testing today, I've seen ping response time as high as 2000 ms! Let's ignore the application level if we can. I believe this network hardware and environmental conditions are capable of much lower latency as confirmed by other posts in this forum. More details of why I'm focused solely on network latency are below if you'd like details.

What can I do, if anything, to decrease latency, or better tune this link for high frequency requests for small amounts of data over SMB?

Why network latency is my focus:
I have proven latency to be the only issue. Client is requesting LOTS of tiny slices (1kb or less) of data from large db tables (20-100Mb) stored as files on SMB. Throughput is not an issue, there's plenty of room left on the line as proven by Btest.exe to RB, and raw SMB transfers of entire large files using same client and server (rather than tons of tiny data requests the db app is issuing).

I've analyzed raw packets and confirmed SMB session is stable, with no re-authentication or obvious signs of issues.

There is no db server component to execute optimized queries, so the client is scanning tables itself. Further, I don't think there's any table indexes, etc to help negate the lack of db server component. I'd go so far as to say the report queries are likely pulling more data than is actually displayed. But, I can't replace the application, nor migrate it to a server/client db engine, nor optimize queries, nor add indices.

If i were in your shoes i would be attempting to narrow the channel width down to 20mhz (you are unlikely to find 40mhz in the 2Ghz band); limit your MCS rates to 0->13 and run nstreme is your protocol. If this all goes to plan you should easily get 60mbit one way with 0-1ms latency.