You can easily disable the protection of DoS attacks on a Windows server. However, this will not necessarily solve your problem. For 100,000 connections, you will need several servers for something like this.
Your first limitation is that for each IP address you have only 65,000 possible TCP ports for serving connections (65,535 but less reserved ports, etc.). Thus, you either need a massive server with several IP addresses (probably an unreliable and single point of failure in your system / application), or you need several servers behind some kind of load balancer.
In addition, with a long poll, you see a sequential “turn” of connections, when each long connection for polling ends and a new one begins. TCP ports are not reused, but rather the fastest configurable TCP wait wait synchronization is 30 seconds. Thus, even 65,000 connections are unrealistic, I would be half as much as reusing the port. Then you need to consider any other HTTP requests coming to this server for web pages, rest api or other static resources. Then consider any other processing that the processors / memory must do to save / format the data. Therefore, I will reduce this, perhaps by another half. I would say that 15,000 clients per server is a realistic maximum. Thus, for 100,000 users, you intend probably 7 servers at least in the balancing cluster.
The last thing I checked, SignalR does not work in a multi-server environment. Similarly, AJAX or any other “frequent update” method will have similar physical restrictions on the number of available TCP ports / sockets, etc. You simply cannot serve 100,000 clients on a single server with a high frequency of HTTP requests like this.
I did a lot of testing with large-scale workloads like WebSync for ASP.net using multiple servers on Amazon EC2. I work at FrozenMountain, and one of my work last year was to do some balancing testing on multiple servers on Amazon EC2 Cloud. Amazon cloud services offer a nice sticky load balancer and easy duplication of your servers for testing. In the "laboratory conditions" (dedicated servers do not do other things), we could exceed 20 thousand customers on a large copy of Amazon , which is a quad-core server with 7.5 GB of RAM.
It is also worth noting that with the latest / IIS / Websync server, you get WebSockets support, which will help reduce port traffic and reuse, as each client can support one permanent socket connection on the web server. This could potentially increase the number of your clients per server. Thus, you can switch from the 7th server cluster to 4-5 servers depending on the speed of implementation of websocket-compatible browsers / clients. (Clients based on web browsers will not have high adoption speeds, native devices such as iPhone or Android devices will all have web support, so you will immediately see all the benefits). WebSync cannot return from WebSockets to Long Polling if the client does not support web sockets.