NETWORK STRESS TESTING
Stress testing network equipment before operational deployment is a fundamental step to ensure the quality of network services and avoid surprises. Today’s complex network scenarios often involve equipment maintaining state (at L3/4 or at application level). Only generating stateful, realistic application mixes one can be confident that tested equipments will perform equally well in operation.
Traditionally, network stress test generators were either relying on dedicated HW (ASIC, FPGA, NPU) and very expensive, or relying on standard HW but very limited in performances.
The INTEL® DPDK software libraries is a framework that improved packet processing speeds on standard HW up by 10 times, thus enabling high speed, feature-rich, application level traffic generators at affordable costs.
Stress test traffic generators were either HW based and very expensive or SW based and poor in performances
high speed application emulation at affordable costs
DITGBox is a traffic generator based on standard HW only that emulates real application traffic for a wide range of protocols and applications.
DITGBox can either craft packets at L3 or L4 and statistically control per flow inter-packet times and packet lengths (so that the output emulates a traffic mix with some wanted characteristics) or directly generate at L7 with full stack traversal so that application state is maintained as if packets were generated by thousands of separate end systems. Moreover, DITGBox can replay pre-collected real traffic traces either at full link speed (up to 100 Gbps from a 2U form factor server) or modulating the original packets timing.
DITGBox Exploits the INTEL® DPDK framework and uses only standard Hardware components
DITGBox has a very realistic emulation of many applications
DITGBox can either act as a pure traffic sender (“open loop” configuration) or be both a sender and a receiver (“loopback” configuration), or send to different peer DITGBoxes that will act as receivers and potentially send back some traffic (“2-point” configuration, see picture on the left).
In all cases DITGBox is piloted via an easy to use GUI (see below) that graphically maps the equipment’s backplane showing the status of all interfaces, and allows configuring, running, modifying and repeating experiments.
Performances measured by DITGBox are about the traffic mix as a whole (e.g. the packet loss, latency and jitter introduced by an equiment under test) or about the QoS experienced by specific flows within the generated traffic mix.
packet traces tunable playback
DITGBox can work as a flexible playback of pre-collected packet traces. The first working mode is the “full speed trace replica” that plays back a trace at the full speed of an interface (up to 40Gbit/s can be supported). The second working mode is a “tunable trace replica” that selects specific flows from a trace and plays them back varying some characteristics (e.g. L3/4 fields, payload). Each selected flow can be multiplied N times and be played back in parallel, thus emulating N users of the flow’s application.
The tunable trace replica mode is useful to measure the QoS received by specific applications within a complex traffic mix
Custom traffic generation
The third working mode of DITGBox is the Custom Traffic Generation, where characteristics of the packets to be generated can be defined at at all layers of the protocol stack. A wide pre-defined application library is available to simplify this process. The generation can be further tuned by applying statistical models to some elements of the packet generation process (e.g. applying a specific distribution to inter-packet times, or to packet sizes).
Custom Traffic Generation is useful to emulate normal or abnormal traffic conditions (e.g. massive DDoS attacks)
DITGBox can saturate equipment and network portions with realistic traffic mixes, and measure their performance under stress