traccar on Windows server - tuning windows

I have traccar 4.6 setup on windows server 2016

I'm seeing an issue where suddenly the data stops coming in (well some does come in but very little)
The devices report every 1 to 5 secs (not my decision!)
I've tried playing with the server.timeout setting between 6 secs to 1 min 30 but same issue persists

There does seem to be a pattern where netstat show high established and close_wait connections
when the system is working on, its sitting around 2800 established and 500 close wait

But i've seen it report as 30,000 established / close wait, and that's where things go wrong

I was wondering is there some windows settings i need to change to make it work well with traccar?
I've editing the registry to set max TCP connections to 65,000, but not sure what else to try

30k connections is a very high number. How many devices do you have? How often do they report data?

The report rate is high (out of my control)
Every 5 secs, but lots are reporting every second
Its around 4,000 devices

Maybe you need to configure something like this:

https://technet.microsoft.com/en-us/library/cc758980(v=ws.10).aspx

But you would probably need to do some research. I only use Linux for our servers.

Thanks Anton
If that works i'll let you know

But I'm thinking I'll need to switch to use something like Ubuntu

Thanks Anton, I did a bunch of profiling, it turned out to be IO bound issue based on Database living on different server.
When moving the database locally the problem went away.

Out of interest can disabling writing positions/events to database be disabled ?

Not without code changes.

Thanks Anton
I ran some tests earlier with not saving to Db (code changes)

it was easily running with 3,000 messages a sec, but when local database turned on it struggled, and with external database on it was unusable

One interesting observation is that the netstat established connections was reading very high with this external database, not sure if that just weird windows thing/completely incorrect or IO saturation causing side effects

Anyway, thanks for your input, scaling up and local DB will fix the issue