Authorized Network Testing
The traffic matches what an attacker sends. The difference is that you can prove the target is yours and keep the record afterwards.
Authorized testing only - you must own or have explicit permission to test.
Packets per second, connections per second, p95 latency and error rate. Confuse those four and the test is misread. 200,000 packets per second is not the same event as 20,000 connections per second: the first strains volume, the second strains the state table. Every unit has its own row in the report. The layer 7 stresser tab covers the request side, which is a different queue entirely.
SYN builds half-open handshakes and fills the listen queue. UDP measures raw packet rate and never reaches the application. ACK shows state-table pressure on connections that already exist. Churn tests how fast you can open and close sockets. One method per run: mix two and you are averaging different units, which reads as a number nobody can defend.
A single rate shows today's behaviour, not capacity. For a curve, raise the rate in steps and write down p95 at each one. The break point is the step where the rate starts falling back, and it belongs to the target's own ceiling rather than your plan limit. The same target measured on another day adds a second point, and enough points close the curve.
The network runs at 4 Tbps, and VIP plans add dedicated servers that push one test to 50-60 Gbit/s. Plans differ in concurrency and window length rather than methods: Basic runs one test, Pro runs three, and a window reaches 5000 seconds. Sending 50 Gbit/s at an endpoint on a 1 Gbit/s uplink measures your own connection, not ours. Check the uplink before you size the run.
The target is recorded when you add it and checked again the second the run starts. Traffic never leaves for an unconfirmed target and no credit is spent. There is no switch to disable that check, and the stresser panel does not carry one. If it did, the product would be an IP booter with better typography.
The packet path can be healthy while the application collapses. Connection pools fill, database latency climbs, response time crosses 500 ms, and the network sees none of it because packets arrive on time. That is the moment to change layer rather than raise volume, and the free L7 stresser shows the same target from the request side without a card.
A short run gives a point, a long run gives a trend. Warm-up, pool filling and cache behaviour settle in the first minutes, and a short window never sees them settle. A long window can leave a mark on the target, so measure recovery time afterwards too. Do not run it while production traffic is flowing; the result becomes two events blended into one.
The report says where the target fell, not why. Application logs, slow queries and kernel counters are not in it and should not be. Those stay on your side, and the answer usually appears when you put the report's p95 next to a timestamp from your own log. The report alone is a measurement; next to the log it becomes a diagnosis.
On infrastructure you own or have written permission to test, yes. Pointed anywhere else it is a computer crime in most countries.
SYN, UDP, ACK and connection churn, each metered in its own unit.
The network has 4 Tbps, VIP plans push a single test to 50-60 Gbit/s, and your plan sets the ceiling the form enforces.
Short on the free tier, up to 5000 seconds on paid plans.
No. A $0 Basic account or trial credit from live support is enough, and neither needs a card.
Packets and connections per second, p50/p95/p99, error rate and the configuration the run used.
Trial credit from live support or the $0 Basic plan, same panel and same method list. The report matches what the paid plans produce.