If a new router doesn't solve the issue and the latency spikes are because the NAT table in the router gets overwhelmed by too many connections from different IPs, what I may have to do is use an external server as a web server and then route all requests to the router here (so
Conversation
it is the same IP each time -- I can forward the source IP via a proxy header). We're getting around 5k to 10k requests per second and they come from thousands of different IPs -- so my running theory is that the router's NAT table overflows and it has to internally reset
1
4
something to accomodate all the IP addresses. If I did that, it would probably add 50-100ms for every call which isn't ideal. Ideally I can find a Verizon router that is enterprise class.
Replying to
I think you're probably right, this sounds like the NAT table filling up. I've dealt with this problem before exactly as you described; frontend webserver proxies all the requests. (In my case it also goes across a VPN to the backend)