Control flow based cost analysis for P4

Abstract

The networking industry is currently undergoing a steady trend of softwarization. Yet, network engineers suffer from the lack of software development tools that support programming of new protocols. We are creating a cost analysis tool for the P4 programming language, that automatically verifies whether the developed program meets soft deadline requirements imposed by the network. In this paper, we present an approach to estimate the average execution time of P4 program based on control flow graphs. Our approach takes into consideration that many of the parts of P4 are implementation-defined: required information can be added in through incremental refinement, while missing information is handled by falling back to less precise defaults. We illustrate application of this approach to a P4 protocol in two case studies: we use it to examine the effect of a compiler optimization in the deparse stage, and to show how it enables cost modelling complex lookup table implementations. Finally, we assess future research tasks to be completed before the tool is ready for real-world usage.

Model Checking-Based Performance Prediction for P4

“…An important component of our cost analysis tool is the PRISM probabilistic model checker [9]. We chose model checking over common static analysis methods (e.g., control flow analysis, which was examined in our earlier work [12]) because these methods usually do not understand program semantics fully and cannot take into account changes in the program state entirely. The reason we chose probabilistic reward-based model checking in particular over other forms of model checking (for example, UPPAAL, which is built around timed automata and used for the verification of real-time requirements (e.g., in [7])) was due to the large number of unknown factors in the case of P4.…”

“…(We also need to know how table lookups are performed. We dealt with this topic in [12], where we presented a probabilistic (Markov) model of the DIR-24-8 algorithm.) For this reason, we compile P4 down to the PRISM probabilistic model checker which, given the appropriate operational semantics of P4, performs probabilistic cost analysis out of the box, taking into account the side effects and program control flow as well.…”

“…Note that in Section 2, we limited the probabilistically modeled phenomena to input packets (I), while we used statistics ( θ Y ) to address other environment-related unknowns (E). Yet, with the above mechanism, we could easily expand the probabilistically modeled factors (and shrink E at the same time) as well, which allows ample room for future work (for example, to include the probabilistic model of DIR-24-8 in [12]).…”

“…To model these paths, we should model lookup algorithms in PRISM as described in Section 3. While we omitted this step from our current work (and in this experiment, we use tables that never match), in [12], we already modeled the DIR-24-8 lookup algorithm probabilistically in the form of a relatively small Markov chain. This can be directly implemented in PRISM as well.…”

“…Understanding the interplay between protocols and network traffic is useful for optimizing the pipeline. In [12], we examined how the effectiveness of a compiler optimization step-found in the T4P4S compiler-depends on the probability that the size of packet headers is equal to that emitted by a given protocol. Using our current program representation, we can formalize such requirements in PRISM as PCTL formulas that check such probabilities.…”