We provide our services «as is», we don't guarantee target's downtime. Power for each concurrent is dedicated (for example, 5 concurrents deliver 5 times more pps/rqps than 1). We, on the other hand, use as low as 2-3 rqps per each IP by default, and can go as low as 0.1 rqps/IP to bypass extreme ratelimits and still overload the webserver.
Socket Flood is a method that is very easy to filter (and it uses many rqps for each IP!) and its requests are extremely lightweight and don't produce decent load on a server making it great for showing 50000 rqps on dstat but weak/ineffective against actual targets. That said, we guarantee a minimum of 300k+ pps per concurrent for Layer 4 (the Gbps will depend on the method) and minimum 350rqps of valid, full HTTP requests(usually much more at around 1750-5500 rqps) which produce high load on webserver with low request count, unlike 'Socket HTTP Flood' method advertised by majority of other stressers. We also don't want to do false advertising and giving exaggerated or misleading numbers (as many do).
Giving specific numbers is very hard as they depend on many factors, such as where it was tested and the network status at that moment, among others.
