Like last year we are organising new, experimental, cloud and parallel tracks for SMT-COMP. Similar tracks were organised in the SAT competition 2020 and the competition had a positive impact on the development of parallel SAT solvers (see https://satcompetition.github.io/2020/).
The goal in both the cloud and the parallel tracks is to measure the success of a solver in solving a single, hard instance. This will be done by giving solvers instances one at a time, similar to the SMT-COMP single-query track. The participating solvers will be scored based on the number of instances that a solver solves within the per-instance wall-clock time limit and the total run time, similar to the single-query track’s parallel score.
For these tracks we need to choose in total 400 benchmarks from the single-query track logics, and we are specifically reaching out to you, the community and especially the competitors, for suggesting suitable instances to be included in the tracks. In addition we will try to identify instances that are considered interesting. All instances should come from the smt-lib benchmark library.
The solver submission rules follow those of the rest of the tracks. However, on this track we do accept portfolio solvers, as defined in the rules, as competitors. We encourage submission of non-portfolio solvers and reserve the right to give special mentions to non-portfolio solvers in reporting the results.
While the standard competition will be run in StarExec, the parallel and cloud tracks will be run on Amazon Web Services. AWS has kindly agreed to sponsor the participants in the testing phase.
The Cloud Track evaluates the effectiveness of parallel SMT solvers to run in a distributed manner. The solvers participating in this track will be executed with a wall-clock time limit of 20 minutes running on 100 m4.4xlarge machines in parallel. Each m4.4xlarge machine has 16 virtual CPUs and 64 GB memory. Communication between the machines is possible using MPI and SSH.
Participants of this track are required to submit their solver via a GitHub repository (which can be private). The repository should contain a docker file that compiles the solver. As an example, scripts for account configuration and instructions to run HordeSAT in the default configuration are available at https://github.com/aws-samples/aws-batch-comp-infrastructure-sample.
The solvers participating in this track will be executed with a wall-clock time limit of 20 minutes, thus similar to the Single Query Track. Each solver will be run on a single AWS machine of the type m4.16xlarge, which has 64 virtual cores and 256GB of memory. More details about m4.16xlarge nodes can be found here.
Similar to the Cloud Track, participants of this track are required to submit their solver via a GitHub repository (which can be private). The repository should contain a docker file that compiles the solver.
In order to participate in the Cloud or Parallel Track please send an email to email@example.com containing the following: